Opensource ML embedded evaluation kit21.03

Change-Id: I12e807f19f5cacad7cef82572b6dd48252fd61fd

6 files changed, 2857 insertions, 0 deletions

diff --git a/docs/use_cases/ad.md b/docs/use_cases/ad.md
new file mode 100644
index 0000000..ca95af8
--- /dev/null
+++ b/docs/use_cases/ad.md
@@ -0,0 +1,523 @@
+# Anomaly Detection Code Sample
+
+  - [Introduction](#introduction)
+    - [Prerequisites](#prerequisites)
+  - [Building the code sample application from sources](#building-the-code-sample-application-from-sources)
+    - [Build options](#build-options)
+    - [Build process](#build-process)
+    - [Add custom input](#add-custom-input)
+    - [Add custom model](#add-custom-model)
+  - [Setting-up and running Ethos-U55 Code Sample](#setting-up-and-running-ethos-u55-code-sample)
+    - [Setting up the Ethos-U55 Fast Model](#setting-up-the-ethos-u55-fast-model)
+    - [Starting Fast Model simulation](#starting-fast-model-simulation)
+    - [Running Anomaly Detection](#running-anomaly-detection)
+  - [Anomaly Detection processing information](#anomaly-detection-processing-information)
+    - [Preprocessing and feature extraction](#preprocessing-and-feature-extraction)
+    - [Postprocessing](#postprocessing)
+
+## Introduction
+
+This document describes the process of setting up and running the Arm® Ethos™-U55 Anomaly Detection example.
+
+Use case code could be found in [source/use_case/ad](../../source/use_case/ad]) directory.
+
+### Preprocessing and feature extraction
+
+The Anomaly Detection model that is used with the Code Samples expects audio data to be preprocessed
+in a specific way before performing an inference. This section aims to provide an overview of the feature extraction
+process used.
+
+First the audio data is normalized to the range (-1, 1).
+
+Next, a window of 1024 audio samples are taken from the start of the audio clip. From these 1024 samples we calculate 64
+Log Mel Energies that form part of a Log Mel Spectrogram.
+
+The window is shifted by 512 audio samples and another 64 Log Mel Energies are calculated. This is repeated until we
+have 64 sets of Log Mel Energies.
+
+This 64x64 matrix of values is resized by a factor of 2 resulting in a 32x32 matrix of values.
+
+The average of the training dataset is subtracted from this 32x32 matrix and an inference can then be performed.
+
+We start this process again but shifting the start by 20\*512=10240 audio samples. This keeps repeating until enough
+inferences have been performed to cover the whole audio clip.
+
+### Postprocessing
+
+Softmax is applied to the result of each inference. Based on the machine ID of the wav clip being processed we look at a
+specific index in each output vector. An average of the negative value at this index across all the inferences performed
+for the audio clip is taken. If this average value is greater than a chosen threshold score, then the machine in the
+clip is not behaving anomalously. If the score is lower than the threshold then the machine in the clip is behaving
+anomalously.
+
+### Prerequisites
+
+See [Prerequisites](../documentation.md#prerequisites)
+
+## Building the code sample application from sources
+
+### Build options
+
+In addition to the already specified build option in the main documentation, Anomaly Detection use case adds:
+
+- `ad_MODEL_TFLITE_PATH` - Path to the NN model file in TFLite format. Model will be processed and included into
+the application axf
+    file. The default value points to one of the delivered set of models. Note that the parameters `ad_LABELS_TXT_FILE`,
+    `TARGET_PLATFORM` and `ETHOS_U55_ENABLED` should be aligned with the chosen model, i.e.:
+  - if `ETHOS_U55_ENABLED` is set to `On` or `1`, the NN model is assumed to be optimized. The model will naturally fall
+back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
+  - if `ETHOS_U55_ENABLED` is set to `Off` or `0`, the NN model is assumed to be unoptimized. Supplying an optimized
+model in this case will result in a runtime error.
+
+- `ad_FILE_PATH`: Path to the directory containing audio files, or a path to single WAV file, to be used in the
+    application. The default value points to the resources/ad/samples folder containing the delivered set of audio clips.
+
+- `ad_AUDIO_RATE`: Input data sampling rate. Each audio file from ad_FILE_PATH is preprocessed during the build to match
+NN model input requirements.
+    Default value is 16000.
+
+- `ad_AUDIO_MONO`: If set to ON the audio data will be converted to mono. Default is ON.
+
+- `ad_AUDIO_OFFSET`: Start loading audio data starting from this offset (in seconds). Default value is 0.
+
+- `ad_AUDIO_DURATION`: Length of the audio data to be used in the application in seconds. Default is 0 meaning the
+    whole audio file will be taken.
+
+- `ad_AUDIO_MIN_SAMPLES`: Minimum number of samples required by the network model. If the audio clip is shorter than
+    this number, it is padded with zeros. Default value is 16000.
+
+- `ad_MODEL_SCORE_THRESHOLD`: Threshold value to be applied to average softmax score over the clip, if larger than this
+score we have an anomaly.
+
+- `ad_ACTIVATION_BUF_SZ`: The intermediate/activation buffer size reserved for the NN model. By default, it is set to
+    2MiB and should be enough for most models.
+
+In order to build **ONLY** Anomaly Detection example application add to the `cmake` command line specified in [Building](../documentation.md#Building) `-DUSE_CASE_BUILD=ad`.
+
+### Build process
+
+> **Note:** This section describes the process for configuring the build for `MPS3: SSE-300` for different target
+>platform see [Building](../documentation.md#Building).
+
+Create a build directory folder and navigate inside:
+
+```commandline
+mkdir build_ad && cd build_ad
+```
+
+On Linux, execute the following command to build **only** Anomaly Detection application to run on the Ethos-U55 Fast Model when providing only the mandatory arguments for CMake configuration:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+For Windows, add `-G "MinGW Makefiles"`:
+
+```commandline
+cmake \
+    -G "MinGW Makefiles" \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+Toolchain option `CMAKE_TOOLCHAIN_FILE` points to the toolchain specific file to set the compiler and platform specific
+parameters.
+
+To configure a build that can be debugged using Arm-DS, we can just specify
+the build type as `Debug`:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+To configure a build that can be debugged using a tool that only supports
+DWARF format 3 (Modeldebugger for example), we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DARMCLANG_DEBUG_DWARF_LEVEL=3 \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+> **Note:** If building for different Ethos-U55 configurations, see
+[Configuring build for different Arm Ethos-U55 configurations](../sections/building.md#Configuring-build-for-different-Arm-Ethos-U55-configurations):
+
+If the TensorFlow source tree is not in its default expected location,
+set the path using `TENSORFLOW_SRC_PATH`.
+Similarly, if the Ethos-U55 driver is not in the default location,
+`ETHOS_U55_DRIVER_SRC_PATH` can be used to configure the location. For example:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=/my/custom/location/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=/my/custom/location/core_driver \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+Also, `CMSIS_SRC_PATH` parameter can be used to override the CMSIS sources used for compilation used by TensorFlow by
+default. For example, to use the CMSIS sources fetched by the ethos-u helper script, we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=../ethos-u/core_software/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=../ethos-u/core_software/core_driver \
+    -DCMSIS_SRC_PATH=../ethos-u/core_software/cmsis \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+> **Note:** If re-building with changed parameters values, it is highly advised to clean the build directory and re-run the CMake command.
+
+If the CMake command succeeded, build the application as follows:
+
+```commandline
+make -j4
+```
+
+For Windows, use `mingw32-make`.
+
+Add VERBOSE=1 to see compilation and link details.
+
+Results of the build will be placed under `build/bin` folder:
+
+```tree
+bin
+ ├── ethos-u-.axf
+ ├── ethos-u-ad.htm
+ ├── ethos-u-.map
+ ├── images-ad.txt
+ └── sectors
+      └── ad
+          ├── dram.bin
+          └── itcm.bin
+```
+
+Where:
+
+- `ethos-u-ad.axf`: The built application binary for the Anomaly Detection use case.
+
+- `ethos-u-ad.map`: Information from building the application (e.g. libraries used, what was optimized, location of
+    objects)
+
+- `ethos-u-ad.htm`: Human readable file containing the call graph of application functions.
+
+- `sectors/`: Folder containing the built application, split into files for loading into different FPGA memory regions.
+
+- `Images-ad.txt`: Tells the FPGA which memory regions to use for loading the binaries in sectors/\*\* folder.
+
+### Add custom input
+
+The application anomaly detection on audio data found in the folder, or an individual file, set by the CMake parameter
+``ad_FILE_PATH``.
+
+To run the application with your own audio clips first create a folder to hold them and then copy the custom clips into
+this folder:
+
+```commandline
+mkdir /tmp/custom_files
+
+cp custom_id_00.wav /tmp/custom_files/
+```
+
+> **Note:** The data used for this example comes from
+[https://zenodo.org/record/3384388\#.X6GILFNKiqA](https://zenodo.org/record/3384388\#.X6GILFNKiqA)
+and the model included in this example is trained on the ‘Slider’ part of the dataset.
+The machine ID (00, 02, 04, 06) the clip comes from must be in the file name for the application to work.
+The file name should have a pattern that matches
+e.g. `<any>_<text>_00_<here>.wav` if the audio was from machine ID 00
+or `<any>_<text>_02_<here>.wav` if it was from machine ID 02 etc.
+>
+> **Note:** Clean the build directory before re-running the CMake command.
+
+Next set ad_FILE_PATH to the location of this folder when building:
+
+```commandline
+cmake \
+    -Dad_FILE_PATH=/tmp/custom_files/ \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+The images found in the _DIR folder will be picked up and automatically converted to C++ files during the CMake
+configuration stage and then compiled into the application during the build phase for performing inference with.
+
+The log from the configuration stage should tell you what image directory path has been used:
+
+```log
+-- User option ad_FILE_PATH is set to /tmp/custom_files
+```
+
+After compiling, your custom inputs will have now replaced the default ones in the application.
+
+### Add custom model
+
+The application performs inference using the model pointed to by the CMake parameter ``ad_MODEL_TFLITE_PATH``.
+
+> **Note:** If you want to run the model using Ethos-U55, ensure your custom model has been run through the Vela compiler
+>successfully before continuing. See [Optimize model with Vela compiler](../sections/building.md#Optimize-custom-model-with-Vela-compiler).
+
+An example:
+
+```commandline
+cmake \
+    -Dad_MODEL_TFLITE_PATH=<path/to/custom_ad_model_after_vela.tflite> \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=ad ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+The `.tflite` model file pointed to by `ad_MODEL_TFLITE_PATH` will be converted
+to C++ files during the CMake configuration
+stage and then compiled into the application for performing inference with.
+
+The log from the configuration stage should tell you what model path has been used:
+
+```log
+-- User option TARGET_PLATFORM is set to fastmodel
+-- User option ad_MODEL_TFLITE_PATH is set to <path/to/custom_ad_model_after_vela.tflite>
+...
+-- Using <path/to/custom_ad_model_after_vela.tflite>
+++ Converting custom_ad_model_after_vela.tflite to custom_ad_model_after_vela.tflite.cc
+...
+```
+
+After compiling, your custom model will have now replaced the default one in the application.
+
+ >**Note:** In order to successfully run the model, the NPU needs to be enabled and
+ the platform `TARGET_PLATFORM` is set to mps3 and TARGET_SUBSYSTEM is SSE-200 or SSE-300.
+
+## Setting-up and running Ethos-U55 Code Sample
+
+### Setting up the Ethos-U55 Fast Model
+
+The FVP is available publicly from [Arm Ecosystem FVP downloads
+](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps).
+
+For Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains Ethos-U55 and
+Cortex-M55. The model is currently only supported on Linux based machines. To install the FVP:
+
+- Unpack the archive
+
+- Run the install script in the extracted package
+
+```commandline
+.FVP_Corstone_SSE-300_Ethos-U55.sh
+```
+
+- Follow the instructions to install the FVP to your desired location
+
+### Starting Fast Model simulation
+
+> **Note:** The anomaly detection example does not come pre-built. You will first need to follow the instructions in
+>section 3 for building the application from source.
+
+After building, and assuming the install location of the FVP was set to ~/FVP_install_location, the simulation can be
+started by:
+
+```commandline
+~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55 ./bin/ethos-u-ad.axf
+```
+
+A log output should appear on the terminal:
+
+```log
+telnetterminal0: Listening for serial connection on port 5000
+telnetterminal1: Listening for serial connection on port 5001
+telnetterminal2: Listening for serial connection on port 5002
+telnetterminal5: Listening for serial connection on port 5003
+```
+
+This will also launch a telnet window with the sample application's standard output and error log entries containing
+information about the pre-built application version, TensorFlow Lite Micro library version used, data type as well as
+the input and output tensor sizes of the model compiled into the executable binary.
+
+After the application has started if `ad_FILE_PATH` pointed to a single file (or a folder containing a single input file)
+the inference starts immediately. In case of multiple inputs choice, it outputs a menu and waits for the user input from
+telnet terminal:
+
+```log
+User input required
+Enter option number from:
+
+1. Classify next audio clip
+2. Classify audio clip at chosen index
+3. Run classification on all audio clips
+4. Show NN model info
+5. List audio clips
+
+Choice:
+
+```
+
+1. “Classify next audio clip” menu option will run single inference on the next in line.
+
+2. “Classify audio clip at chosen index” menu option will run inference on the chosen audio clip.
+
+    > **Note:** Please make sure to select audio clip index in the range of supplied audio clips during application build.
+    By default, pre-built application has 4 files, indexes from 0 to 3.
+
+3. “Run ... on all” menu option triggers sequential inference executions on all built-in .
+
+4. “Show NN model info” menu option prints information about model data type, input and output tensor sizes:
+
+    ```log
+    [INFO] uTFL version: 2.5.0
+    [INFO] Model info:
+    [INFO] Model INPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 1024 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:  32
+    [INFO] 		2:  32
+    [INFO] 		3:   1
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.192437
+    [INFO] ZeroPoint[0] = 11
+    [INFO] Model OUTPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 8 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:   8
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.048891
+    [INFO] ZeroPoint[0] = -30
+    [INFO] Activation buffer (a.k.a tensor arena) size used: 198016
+    [INFO] Number of operators: 1
+    [INFO] 	Operator 0: ethos-u
+    [INFO] Use of Arm uNPU is enabled
+
+    ```
+
+5. “List” menu option prints a list of pair ... indexes - the original filenames embedded in the application:
+
+    ```log
+    [INFO] List of Files:
+    [INFO] 0 =>; anomaly_id_00_00000000.wav
+    [INFO] 1 =>; anomaly_id_02_00000076.wav
+    [INFO] 2 =>; normal_id_00_00000004.wav
+    [INFO] 3 =>; normal_id_02_00000001.wav
+    ```
+
+### Running Anomaly Detection
+
+Please select the first menu option to execute Anomaly Detection.
+
+The following example illustrates application output:
+
+```log
+[INFO] Running inference on audio clip 0 => anomaly_id_00_00000000.wav
+[INFO] Inference 1/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1081154
+	Idle NPU cycles:   1012
+
+[INFO] Inference 2/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080934
+	Idle NPU cycles:   232
+
+[INFO] Inference 3/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1081332
+	Idle NPU cycles:   834
+
+[INFO] Inference 4/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080748
+	Idle NPU cycles:   418
+
+[INFO] Inference 5/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080728
+	Idle NPU cycles:   438
+
+[INFO] Inference 6/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1081144
+	Idle NPU cycles:   1022
+
+[INFO] Inference 7/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080924
+	Idle NPU cycles:   242
+
+[INFO] Inference 8/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1081322
+	Idle NPU cycles:   844
+
+[INFO] Inference 9/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080738
+	Idle NPU cycles:   428
+
+[INFO] Inference 10/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080718
+	Idle NPU cycles:   448
+
+[INFO] Inference 11/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1081134
+	Idle NPU cycles:   1032
+
+[INFO] Inference 12/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1080914
+	Idle NPU cycles:   252
+
+[INFO] Inference 13/13
+[INFO] Profile for Inference:
+	Active NPU cycles: 1081312
+	Idle NPU cycles:   854
+
+[INFO] Average anomaly score is: -0.024493
+Anomaly threshold is: -0.800000
+Anomaly detected!
+
+```
+
+As multiple inferences have to be run for one clip it will take around a minute or so for all inferences to complete.
+
+For the anomaly_id_00_00000000.wav clip, after averaging results across all inferences the score is greater than the
+chosen anomaly threshold so an anomaly was detected with the machine in this clip.
+
+The profiling section of the log shows that for each inference. For the last inference the profiling reports:
+
+- Ethos-U55's PMU report:
+
+  - 1,081,312 active cycles: number of cycles that were used for computation
+
+  - 854 idle cycles: number of cycles for which the NPU was idle
+
+- For FPGA platforms, CPU cycle count can also be enabled. For FVP, however, CPU cycle counters should not be used as
+    the CPU model is not cycle-approximate or cycle-accurate.
diff --git a/docs/use_cases/asr.md b/docs/use_cases/asr.md
new file mode 100644
index 0000000..d224aca
--- /dev/null
+++ b/docs/use_cases/asr.md
@@ -0,0 +1,529 @@
+# Automatic Speech Recognition Code Sample
+
+- [Introduction](#introduction)
+  - [Prerequisites](#prerequisites)
+- [Building the code sample application from sources](#building-the-code-sample-application-from-sources)
+  - [Build options](#build-options)
+  - [Build process](#build-process)
+  - [Add custom input](#add-custom-input)
+  - [Add custom model](#add-custom-model)
+- [Setting-up and running Ethos-U55 Code Sample](#setting-up-and-running-ethos-u55-code-sample)
+  - [Setting up the Ethos-U55 Fast Model](#setting-up-the-ethos-u55-fast-model)
+  - [Starting Fast Model simulation](#starting-fast-model-simulation)
+  - [Running Automatic Speech Recognition](#running-automatic-speech-recognition)
+- [Automatic Speech Recognition processing information](#automatic-speech-recognition-processing-information)
+  - [Preprocessing and feature extraction](#preprocessing-and-feature-extraction)
+  - [Postprocessing](#postprocessing)
+
+## Introduction
+
+This document describes the process of setting up and running the Arm® Ethos™-U55 Automatic Speech Recognition example.
+
+Use case code could be found in [source/use_case/asr](../../source/use_case/asr]) directory.
+
+### Preprocessing and feature extraction
+
+The wav2letter automatic speech recognition model that is used with the Code Samples expects audio data to be
+preprocessed in a specific way before performing an inference. This section aims to provide an overview of the feature
+extraction process used.
+
+First the audio data is normalized to the range (-1, 1).
+
+> **Note:** Mel-frequency cepstral coefficients (MFCCs) are a common feature extracted from audio data and can be used as
+>input for machine learning tasks like keyword spotting and speech recognition. See source/application/main/include/Mfcc.hpp
+>for implementation details.
+
+Next, a window of 512 audio samples is taken from the start of the audio clip. From these 512 samples we calculate 13
+MFCC features.
+
+The whole window is shifted to the right by 160 audio samples and 13 new MFCC features are calculated. This process of
+shifting and calculating is repeated until enough audio samples to perform an inference have been processed. In total
+this will be 296 windows that each have 13 MFCC features calculated for them.
+
+After extracting MFCC features the first and second order derivatives of these features with respect to time are
+calculated. These derivative features are then standardized and concatenated with the MFCC features (which also get
+standardized). At this point the input tensor will have a shape of 296x39.
+
+These extracted features are quantized, and an inference is performed.
+
+![ASR preprocessing](../media/ASR_preprocessing.png)
+
+For longer audio clips where multiple inferences need to be performed, then the initial starting position is offset by
+(100*160) = 16000 audio samples. From this new starting point, MFCC and derivative features are calculated as before
+until there is enough to perform another inference. Padding can be used if there are not enough audio samples for at
+least 1 inference. This step is repeated until the whole audio clip has been processed. If there are not enough audio
+samples for a final complete inference the MFCC features will be padded by repeating the last calculated feature until
+an inference can be performed.
+
+> **Note:** Parameters of the MFCC feature extraction such as window size, stride, number of features etc. all depend on
+>what was used during model training. These values are specific to each model. If you switch to a different ASR model
+>than the one supplied, then the feature extraction process could be completely different to the one currently implemented.
+
+The amount of audio samples we offset by for long audio clips is specific to the included wav2letter model.
+
+### Postprocessing
+
+After performing an inference, the raw output need to be postprocessed to get a usable result.
+
+The raw output from the model is a tensor of shape 148x29 where each row is a probability distribution over the possible
+29 characters that can appear at each of the 148 time steps.
+
+This wav2letter model is trained using context windows, this means that only certain parts of the output are usable
+depending on the bit of the audio clip that is currently being processed.
+
+If this is the first inference and multiple inferences are required, then ignore the final 49 rows of the output.
+Similarly, if this is the final inference from multiple inferences then ignore the first 49 rows of the output. Finally,
+if this inference is not the last or first inference then ignore the first and last 49 rows of the model output.
+
+> **Note:** If the audio clip is small enough then the whole of the model output is usable and there is no need to throw
+>away any of the output before continuing.
+
+Once any rows have been removed the final processing can be done. To process the output, first the letter with the
+highest probability at each time step is found. Next, any letters that are repeated multiple times in a row are removed
+(e.g. [t, t, t, o, p, p] becomes [t, o, p]). Finally, the 29th blank token letter is removed from the output.
+
+For the final output, the result from all inferences are combined before decoding. What you are left with is then
+displayed to the console.
+
+### Prerequisites
+
+See [Prerequisites](../documentation.md#prerequisites)
+
+## Building the code sample application from sources
+
+### Build options
+
+In addition to the already specified build option in the main documentation, Automatic Speech Recognition use case
+adds:
+
+- `asr_MODEL_TFLITE_PATH` - Path to the NN model file in TFLite format. Model will be processed and included into the
+application axf file. The default value points to one of the delivered set of models. Note that the parameters
+`asr_LABELS_TXT_FILE`,`TARGET_PLATFORM` and `ETHOS_U55_ENABLED` should be aligned with the chosen model, i.e.:
+  - if `ETHOS_U55_ENABLED` is set to `On` or `1`, the NN model is assumed to be optimized. The model will naturally
+fall back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
+  - if `ETHOS_U55_ENABLED` is set to `Off` or `0`, the NN model is assumed to be unoptimized. Supplying an optimized
+model in this case will result in a runtime error.
+
+- `asr_FILE_PATH`:  Path to the directory containing audio files, or a path to single WAV file, to be used in the
+    application. The default value points
+    to the resources/asr/samples folder containing the delivered set of audio clips.
+
+- `asr_LABELS_TXT_FILE`: Path to the labels' text file. The file is used to map letter class index to the text label.
+    The default value points to the delivered labels.txt file inside the delivery package.
+
+- `asr_AUDIO_RATE`: Input data sampling rate. Each audio file from asr_FILE_PATH is preprocessed during the build to
+    match NN model input requirements. Default value is 16000.
+
+- `asr_AUDIO_MONO`: If set to ON the audio data will be converted to mono. Default is ON.
+
+- `asr_AUDIO_OFFSET`: Start loading audio data starting from this offset (in seconds). Default value is 0.
+
+- `asr_AUDIO_DURATION`: Length of the audio data to be used in the application in seconds. Default is 0 meaning the
+    whole audio file will be taken.
+
+- `asr_AUDIO_MIN_SAMPLES`: Minimum number of samples required by the network model. If the audio clip is shorter than
+    this number, it is padded with zeros. Default value is 16000.
+
+- `asr_MODEL_SCORE_THRESHOLD`: Threshold value that must be applied to the inference results for a label to be
+    deemed valid. Default is 0.5.
+
+- `asr_ACTIVATION_BUF_SZ`: The intermediate/activation buffer size reserved for the NN model. By default, it is set
+    to 2MiB and should be enough for most models.
+
+In order to build **ONLY** automatic speech recognition example application add to the `cmake` command line specified in
+[Building](../documentation.md#Building) `-DUSE_CASE_BUILD=asr`.
+
+### Build process
+
+> **Note:** This section describes the process for configuring the build for `MPS3: SSE-300` for different target
+>platform see [Building](../documentation.md#Building) section.
+
+In order to build **only** the automatic speech recognition example, create a build directory and navigate inside:
+
+```commandline
+mkdir build_asr && cd build_asr
+```
+
+On Linux, execute the following command to build **only** Automatic Speech Recognition application to run on the
+Ethos-U55 Fast Model when providing only the mandatory arguments for CMake configuration:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=asr ..
+```
+
+For Windows, add `-G "MinGW Makefiles"`:
+
+```commandline
+cmake \
+    -G "MinGW Makefiles" \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=asr ..
+```
+
+Toolchain option `CMAKE_TOOLCHAIN_FILE` points to the toolchain specific file to set the compiler and platform specific
+parameters.
+
+To configure a build that can be debugged using Arm-DS, we can just specify
+the build type as `Debug`:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DUSE_CASE_BUILD=asr ..
+```
+
+To configure a build that can be debugged using a tool that only supports
+DWARF format 3 (Modeldebugger for example), we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DARMCLANG_DEBUG_DWARF_LEVEL=3 \
+    -DUSE_CASE_BUILD=asr ..
+```
+
+> **Note:** If building for different Ethos-U55 configurations, see
+>[Configuring build for different Arm Ethos-U55 configurations](../sections/building.md#Configuring-build-for-different-Arm-Ethos-U55-configurations):
+
+If the TensorFlow source tree is not in its default expected location,
+set the path using `TENSORFLOW_SRC_PATH`.
+Similarly, if the Ethos-U55 driver is not in the default location,
+`ETHOS_U55_DRIVER_SRC_PATH` can be used to configure the location. For example:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=/my/custom/location/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=/my/custom/location/core_driver \
+    -DUSE_CASE_BUILD=asr ..
+```
+
+Also, `CMSIS_SRC_PATH` parameter can be used to override the CMSIS sources used for compilation used by TensorFlow by
+default. For example, to use the CMSIS sources fetched by the ethos-u helper script, we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=../ethos-u/core_software/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=../ethos-u/core_software/core_driver \
+    -DCMSIS_SRC_PATH=../ethos-u/core_software/cmsis \
+    -DUSE_CASE_BUILD=asr ..
+```
+
+> **Note:** If re-building with changed parameters values, it is highly advised to clean the build directory and re-run
+>the CMake command.
+
+If the CMake command succeeded, build the application as follows:
+
+```commandline
+make -j4
+```
+
+For Windows, use `mingw32-make`.
+
+Add `VERBOSE=1` to see compilation and link details.
+
+Results of the build will be placed under `build/bin` folder:
+
+```tree
+bin
+ ├── ethos-u-asr.axf
+ ├── ethos-u-asr.htm
+ ├── ethos-u-asr.map
+ ├── images-asr.txt
+ └── sectors
+      └── asr
+          ├── dram.bin
+          └── itcm.bin
+```
+
+Where:
+
+- `ethos-u-asr.axf`: The built application binary for the Automatic Speech Recognition use case.
+
+- `ethos-u-asr.map`: Information from building the application (e.g. libraries used, what was optimized, location of
+    objects)
+
+- `ethos-u-asr.htm`: Human readable file containing the call graph of application functions.
+
+- `sectors/`: Folder containing the built application, split into files for loading into different FPGA memory regions.
+
+- `Images-asr.txt`: Tells the FPGA which memory regions to use for loading the binaries in sectors/** folder.
+
+### Add custom input
+
+The application performs inference on audio data found in the folder, or an individual file, set by the CMake parameter
+`asr_FILE_PATH`.
+
+To run the application with your own audio clips first create a folder to hold them and then copy the custom audio clips
+into this folder:
+
+```commandline
+mkdir /tmp/custom_wavs
+
+cp my_clip.wav /tmp/custom_wavs/
+```
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+Next set `asr_FILE_PATH` to the location of this folder when building:
+
+```commandline
+cmake \
+    -Dasr_FILE_PATH=/tmp/custom_wavs/ \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DUSE_CASE_BUILD=asr \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+The audio clips found in the `asr_FILE_PATH` folder will be picked up and automatically converted to C++ files during the
+CMake configuration stage and then compiled into the application during the build phase for performing inference with.
+
+The log from the configuration stage should tell you what audio clip directory path has been used:
+
+```log
+-- User option asr_FILE_PATH is set to /tmp/custom_wavs
+-- Generating audio files from /tmp/custom_wavs
+++ Converting my_clip.wav to my_clip.cc
+++ Generating build/generated/asr/include/InputFiles.hpp
+++ Generating build/generated/asr/src/InputFiles.cc
+-- Defined build user options:
+-- asr_FILE_PATH=/tmp/custom_wavs
+```
+
+After compiling, your custom inputs will have now replaced the default ones in the application.
+
+> **Note:** The CMake parameter asr_AUDIO_MIN_SAMPLES determine the minimum number of input sample. When building the
+>application, if the size of the audio clips is less then asr_AUDIO_MIN_SAMPLES then it will be padded so that it does.
+
+### Add custom model
+
+The application performs inference using the model pointed to by the CMake parameter MODEL_TFLITE_PATH.
+
+> **Note:** If you want to run the model using Ethos-U55, ensure your custom model has been run through the Vela
+>compiler successfully before continuing. See [Optimize model with Vela compiler](../sections/building.md#Optimize-custom-model-with-Vela-compiler).
+
+To run the application with a custom model you will need to provide a labels_<model_name>.txt file of labels
+associated with the model. Each line of the file should correspond to one of the outputs in your model. See the provided
+labels_wav2letter.txt file for an example.
+
+Then, you must set `asr_MODEL_TFLITE_PATH` to the location of the Vela processed model file and `asr_LABELS_TXT_FILE`to
+the location of the associated labels file.
+
+An example:
+
+```commandline
+cmake \
+    -Dasr_MODEL_TFLITE_PATH=<path/to/custom_model_after_vela.tflite> \
+    -Dasr_LABELS_TXT_FILE=<path/to/labels_custom_model.txt> \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+The `.tflite` model file pointed to by `asr_MODEL_TFLITE_PATH` and labels text file pointed to by `asr_LABELS_TXT_FILE`
+will be converted to C++ files during the CMake configuration stage and then compiled into the application for performing
+inference with.
+
+The log from the configuration stage should tell you what model path and labels file have been used:
+
+```log
+-- User option TARGET_PLATFORM is set to mps3
+-- User option asr_MODEL_TFLITE_PATH is set to <path/to/custom_model_after_vela.tflite>
+...
+-- User option asr_LABELS_TXT_FILE is set to <path/to/labels_custom_model.txt>
+...
+-- Using <path/to/custom_model_after_vela.tflite>
+++ Converting custom_model_after_vela.tflite to\
+custom_model_after_vela.tflite.cc
+-- Generating labels file from <path/to/labels_custom_model.txt>
+-- writing to <path/to/build/generated/src/Labels.cc>
+...
+```
+
+After compiling, your custom model will have now replaced the default one in the application.
+
+## Setting-up and running Ethos-U55 Code Sample
+
+### Setting up the Ethos-U55 Fast Model
+
+The FVP is available publicly from [Arm Ecosystem FVP downloads
+](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps).
+
+For Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains Ethos-U55 and
+Cortex-M55. The model is currently only supported on Linux based machines. To install the FVP:
+
+- Unpack the archive
+
+- Run the install script in the extracted package
+
+```commandline
+./FVP_Corstone_SSE-300_Ethos-U55.sh
+```
+
+- Follow the instructions to install the FVP to your desired location
+
+### Starting Fast Model simulation
+
+Once completed the building step, application binary ethos-u-asr.axf can be found in the `build/bin` folder.
+Assuming the install location of the FVP was set to ~/FVP_install_location, the simulation can be started by:
+
+```commandline
+~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55
+./bin/mps3-sse-300/ethos-u-asr.axf
+```
+
+A log output should appear on the terminal:
+
+```log
+telnetterminal0: Listening for serial connection on port 5000
+telnetterminal1: Listening for serial connection on port 5001
+telnetterminal2: Listening for serial connection on port 5002
+telnetterminal5: Listening for serial connection on port 5003
+```
+
+This will also launch a telnet window with the sample application's standard output and error log entries containing
+information about the pre-built application version, TensorFlow Lite Micro library version used, data type as well as
+the input and output tensor sizes of the model compiled into the executable binary.
+
+After the application has started if `asr_FILE_PATH` pointed to a single file (or a folder containing a single input file)
+the inference starts immediately. In case of multiple inputs choice, it outputs a menu and waits for the user input from
+telnet terminal:
+
+```log
+User input required
+Enter option number from:
+
+1. Classify next audio clip
+2. Classify audio clip at chosen index
+3. Run classification on all audio clips
+4. Show NN model info
+5. List audio clips
+
+Choice:
+
+```
+
+1. “Classify next audio clip” menu option will run inference on the next in line voice clip from the collection of the
+    compiled audio.
+
+    > **Note:** Note that if the clip is over a certain length, the application will invoke multiple inference runs to
+>cover the entire file.
+
+2. “Classify audio clip at chosen index” menu option will run inference on the chosen audio clip.
+
+    > **Note:** Please make sure to select audio clip index in the range of supplied audio clips during application build.
+    By default, pre-built application has 4 files, indexes from 0 to 3.
+
+3. “Run classification on all audio clips” menu option triggers sequential inference executions on all built-in voice
+    samples.
+
+4. “Show NN model info” menu option prints information about model data type, input and output tensor sizes:
+
+    ```log
+    [INFO] uTFL version: 2.5.0
+    [INFO] Model info:
+    [INFO] Model INPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 11544 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1: 296
+    [INFO] 		2:  39
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.110316
+    [INFO] ZeroPoint[0] = -11
+    [INFO] Model OUTPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 4292 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:   1
+    [INFO] 		2: 148
+    [INFO] 		3:  29
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.003906
+    [INFO] ZeroPoint[0] = -128
+    [INFO] Activation buffer (a.k.a tensor arena) size used: 783168
+    [INFO] Number of operators: 1
+    [INFO] 	Operator 0: ethos-u
+    [INFO] Use of Arm uNPU is enabled
+    ```
+
+5. “List” menu option prints a list of pair audio clip indexes - the original filenames embedded in the application:
+
+    ```log
+    [INFO] List of Files:
+    [INFO] 0 => anotherdoor.wav
+    [INFO] 1 => anotherengineer.wav
+    [INFO] 2 => itellyou.wav
+    [INFO] 3 => testingroutine.wav
+    ```
+
+### Running Automatic Speech Recognition
+
+Please select the first menu option to execute Automatic Speech Recognition.
+
+The following example illustrates application output:
+
+```log
+[INFO] Running inference on audio clip 0 => anotherdoor.wav
+[INFO] Inference 1/2
+[INFO] Profile for pre-processing:
+	Active NPU cycles: 0
+	Idle NPU cycles:   6
+
+[INFO] Profile for Inference:
+	Active NPU cycles: 28924342
+	Idle NPU cycles:   824
+
+[INFO] Inference 2/2
+[INFO] Profile for pre-processing:
+	Active NPU cycles: 0
+	Idle NPU cycles:   6
+
+[INFO] Profile for Inference:
+	Active NPU cycles: 28924298
+	Idle NPU cycles:   868
+
+[INFO] Result for inf 0: and he walked immediately out o t
+[INFO] Result for inf 1: he aparctment by anoer dor
+[INFO] Final result: and he walked immediately out o the aparctment by anoer dor
+```
+
+It could take several minutes to complete each inference (average time is 5-7 minutes), and on this audio clip multiple
+inferences were required to cover the whole clip.
+
+The profiling section of the log shows that for the first inference:
+
+- Ethos-U55's PMU report:
+
+  - 28,924,298 active cycles: number of NPU cycles that were used for computation
+
+  - 868 idle cycles: number of cycles for which the NPU was idle
+
+- For FPGA platforms, CPU cycle count can also be enabled. For FVP, however, CPU cycle counters should not be used as
+    the CPU model is not cycle-approximate or cycle-accurate.
+
+The application prints the decoded output from each of the inference runs as well as the final combined result.
diff --git a/docs/use_cases/img_class.md b/docs/use_cases/img_class.md
new file mode 100644
index 0000000..7a409f2
--- /dev/null
+++ b/docs/use_cases/img_class.md
@@ -0,0 +1,446 @@
+# Image Classification Code Sample
+
+- [Introduction](#introduction)
+  - [Prerequisites](#prerequisites)
+- [Building the code sample application from sources](#building-the-code-sample-application-from-sources)
+  - [Build options](#build-options)
+  - [Build process](#build-process)
+  - [Add custom input](#add-custom-input)
+  - [Add custom model](#add-custom-model)
+- [Setting-up and running Ethos-U55 code sample](#setting-up-and-running-ethos-u55-code-sample)
+  - [Setting up the Ethos-U55 Fast Model](#setting-up-the-ethos-u55-fast-model)
+  - [Starting Fast Model simulation](#starting-fast-model-simulation)
+  - [Running Image Classification](#running-image-classification)
+
+## Introduction
+
+This document describes the process of setting up and running the Arm® Ethos™-U55 Image Classification
+example.
+
+Use case solves classical computer vision problem: image classification. The ML sample was developed using MobileNet v2
+model trained on ImageNet dataset.
+
+Use case code could be found in [source/use_case/img_class](../../source/use_case/img_class]) directory.
+
+### Prerequisites
+
+See [Prerequisites](../documentation.md#prerequisites)
+
+## Building the code sample application from sources
+
+### Build options
+
+In addition to the already specified build option in the main documentation, Image Classification use case specifies:
+
+- `img_class_MODEL_TFLITE_PATH` - Path to the NN model file in TFLite format. Model will be processed and included into
+    the application axf file. The default value points to one of the delivered set of models. Note that the parameters
+    `img_class_LABELS_TXT_FILE`,`TARGET_PLATFORM` and `ETHOS_U55_ENABLED` should be aligned with the chosen model, i.e.:
+  - if `ETHOS_U55_ENABLED` is set to `On` or `1`, the NN model is assumed to be optimized. The model will naturally
+    fall back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
+  - if `ETHOS_U55_ENABLED` is set to `Off` or `0`, the NN model is assumed to be unoptimized. Supplying an optimized
+    model in this case will result in a runtime error.
+
+- `img_class_FILE_PATH`: Path to the directory containing images, or path to a single image file, to be used file(s) in
+    the application. The default value points to the resources/img_class/samples folder containing the delivered
+    set of images. See more in the [Add custom input data section](#add-custom-input).
+
+- `img_class_IMAGE_SIZE`: The NN model requires input images to be of a specific size. This parameter defines the
+    size of the image side in pixels. Images are considered squared. Default value is 224, which is what the supplied
+    MobilenetV2-1.0 model expects.
+
+- `img_class_LABELS_TXT_FILE`: Path to the labels' text file to be baked into the application. The file is used to
+    map classified classes index to the text label. Change this parameter to point to the custom labels file to map
+    custom NN model output correctly.\
+    The default value points to the delivered labels.txt file inside the delivery package.
+
+- `img_class_ACTIVATION_BUF_SZ`: The intermediate/activation buffer size reserved for the NN model. By default, it
+    is set to 2MiB and should be enough for most models.
+
+- `USE_CASE_BUILD`: set to img_class to build only this example.
+
+In order to build **ONLY** Image Classification example application add to the `cmake` command line specified in
+[Building](../documentation.md#Building) `-DUSE_CASE_BUILD=img_class`.
+
+### Build process
+
+> **Note:** This section describes the process for configuring the build for `MPS3: SSE-300` for different target platform
+see [Building](../documentation.md#Building).
+
+Create a build directory folder and navigate inside:
+
+```commandline
+mkdir build_img_class && cd build_img_class
+```
+
+On Linux, execute the following command to build **only** Image Classification application to run on the Ethos-U55 Fast
+Model when providing only the mandatory arguments for CMake configuration:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+For Windows, add `-G "MinGW Makefiles"`:
+
+```commandline
+cmake \
+    -G "MinGW Makefiles" \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+Toolchain option `CMAKE_TOOLCHAIN_FILE` points to the toolchain specific file to set the compiler and platform specific
+parameters.
+
+To configure a build that can be debugged using Arm-DS, we can just specify
+the build type as `Debug`:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+To configure a build that can be debugged using a tool that only supports
+DWARF format 3 (Modeldebugger for example), we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DARMCLANG_DEBUG_DWARF_LEVEL=3 \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+> **Note:** If building for different Ethos-U55 configurations, see
+>[Configuring build for different Arm Ethos-U55 configurations](../sections/building.md#Configuring-build-for-different-Arm-Ethos-U55-configurations):
+
+If the TensorFlow source tree is not in its default expected location,
+set the path using `TENSORFLOW_SRC_PATH`.
+Similarly, if the Ethos-U55 driver is not in the default location,
+`ETHOS_U55_DRIVER_SRC_PATH` can be used to configure the location. For example:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=/my/custom/location/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=/my/custom/location/core_driver \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+Also, `CMSIS_SRC_PATH` parameter can be used to override the CMSIS sources used for compilation used by TensorFlow by
+default. For example, to use the CMSIS sources fetched by the ethos-u helper script, we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=../ethos-u/core_software/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=../ethos-u/core_software/core_driver \
+    -DCMSIS_SRC_PATH=../ethos-u/core_software/cmsis \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+> **Note:** If re-building with changed parameters values, it is highly advised to clean the build directory and re-run
+>the CMake command.
+
+If the CMake command succeeds, build the application as follows:
+
+```commandline
+make -j4
+```
+
+For Windows, use `mingw32-make`.
+
+Add VERBOSE=1 to see compilation and link details.
+
+Results of the build will be placed under `build/bin` folder:
+
+```tree
+bin
+ ├── ethos-u-img_class.axf
+ ├── ethos-u-img_class.htm
+ ├── ethos-u-img_class.map
+ ├── images-img_class.txt
+ └── sectors
+      └── img_class
+           ├── dram.bin
+           └── itcm.bin
+```
+
+Where:
+
+- `ethos-u-img_class.axf`: The built application binary for the Image Classification use case.
+
+- `ethos-u-img_class.map`: Information from building the application (e.g. libraries used, what was optimized, location
+    of objects)
+
+- `ethos-u-img_class.htm`: Human readable file containing the call graph of application functions.
+
+- `sectors/`: Folder containing the built application, split into files for loading into different FPGA memory regions.
+
+- `Images-img_class.txt`: Tells the FPGA which memory regions to use for loading the binaries in sectors/** folder.
+
+### Add custom input
+
+The application performs inference on input data found in the folder, or an individual file set by the CMake parameter
+img_class_FILE_PATH.
+
+To run the application with your own images, first create a folder to hold them and then copy the custom images into
+this folder, for example:
+
+```commandline
+mkdir /tmp/custom_images
+
+cp custom_image1.bmp /tmp/custom_images/
+```
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+Next set `img_class_FILE_PATH` to the location of this folder when building:
+
+```commandline
+cmake \
+    -Dimg_class_FILE_PATH=/tmp/custom_images/ \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+The images found in the `img_class_FILE_PATH` folder will be picked up and automatically converted to C++ files during
+the CMake configuration stage and then compiled into the application during the build phase for performing inference
+with.
+
+The log from the configuration stage should tell you what image directory path has been used:
+
+```log
+-- User option img_class_FILE_PATH is set to /tmp/custom_images
+-- User option img_class_IMAGE_SIZE is set to 224
+...
+-- Generating image files from /tmp/custom_images
+++ Converting custom_image1.bmp to custom_image1.cc
+...
+-- Defined build user options:
+...
+-- img_class_FILE_PATH=/tmp/custom_images
+-- img_class_IMAGE_SIZE=224
+```
+
+After compiling, your custom images will have now replaced the default ones in the application.
+
+> **Note:** The CMake parameter IMAGE_SIZE should match the model input size. When building the application,
+if the size of any image does not match IMAGE_SIZE then it will be rescaled and padded so that it does.
+
+### Add custom model
+
+The application performs inference using the model pointed to by the CMake parameter MODEL_TFLITE_PATH.
+
+> **Note:** If you want to run the model using Ethos-U55, ensure your custom model has been run through the Vela compiler
+>successfully before continuing. See [Optimize model with Vela compiler](../sections/building.md#Optimize-custom-model-with-Vela-compiler).
+
+To run the application with a custom model you will need to provide a labels_<model_name>.txt file of labels
+associated with the model. Each line of the file should correspond to one of the outputs in your model. See the provided
+labels_mobilenet_v2_1.0_224.txt file for an example.
+
+Then, you must set `img_class_MODEL_TFLITE_PATH` to the location of the Vela processed model file and
+`img_class_LABELS_TXT_FILE` to the location of the associated labels file.
+
+An example:
+
+```commandline
+cmake \
+    -Dimg_class_MODEL_TFLITE_PATH=<path/to/custom_model_after_vela.tflite> \
+    -Dimg_class_LABELS_TXT_FILE=<path/to/labels_custom_model.txt> \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=img_class ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+The `.tflite` model file pointed to by `img_class_MODEL_TFLITE_PATH` and labels text file pointed to by
+`img_class_LABELS_TXT_FILE` will be converted to C++ files during the CMake configuration stage and then compiled into
+the application for performing inference with.
+
+The log from the configuration stage should tell you what model path and labels file have been used:
+
+```log
+-- User option img_class_MODEL_TFLITE_PATH is set to <path/to/custom_model_after_vela.tflite>
+...
+-- User option img_class_LABELS_TXT_FILE is set to <path/to/labels_custom_model.txt>
+...
+-- Using <path/to/custom_model_after_vela.tflite>
+++ Converting custom_model_after_vela.tflite to\
+custom_model_after_vela.tflite.cc
+-- Generating labels file from <path/to/labels_custom_model.txt>
+-- writing to <path/to/build/generated/src/Labels.cc>
+...
+```
+
+After compiling, your custom model will have now replaced the default one in the application.
+
+## Setting-up and running Ethos-U55 code sample
+
+### Setting up the Ethos-U55 Fast Model
+
+The FVP is available publicly from [Arm Ecosystem FVP downloads](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps).
+
+For Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains Ethos-U55 and
+Cortex-M55. The model is currently only supported on Linux based machines. To install the FVP:
+
+- Unpack the archive
+
+- Run the install script in the extracted package
+
+```commandline
+$./FVP_Corstone_SSE-300_Ethos-U55.sh
+```
+
+- Follow the instructions to install the FVP to your desired location
+
+### Starting Fast Model simulation
+
+Pre-built application binary ethos-u-img_class.axf can be found in the bin/mps3-sse-300 folder of the delivery package.
+Assuming the install location of the FVP was set to ~/FVP_install_location, the simulation can be started by:
+
+```commandline
+~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55
+./bin/mps3-sse-300/ethos-u-img_class.axf
+```
+
+A log output should appear on the terminal:
+
+```log
+telnetterminal0: Listening for serial connection on port 5000
+telnetterminal1: Listening for serial connection on port 5001
+telnetterminal2: Listening for serial connection on port 5002
+telnetterminal5: Listening for serial connection on port 5003
+```
+
+This will also launch a telnet window with the sample application's standard output and error log entries containing
+information about the pre-built application version, TensorFlow Lite Micro library version used, data type as well as
+the input and output tensor sizes of the model compiled into the executable binary.
+
+After the application has started if `img_class_FILE_PATH` pointed to a single file (or a folder containing a single image)
+the inference starts immediately. In case of multiple inputs choice, it outputs a menu and waits for the user input from
+telnet terminal:
+
+```log
+User input required
+Enter option number from:
+
+1. Classify next image
+2. Classify image at chosen index
+3. Run classification on all images
+4. Show NN model info
+5. List images
+
+Choice:
+
+```
+
+1. “Classify next image” menu option will run single inference on the next in line image from the collection of the
+    compiled images.
+
+2. “Classify image at chosen index” menu option will run single inference on the chosen image.
+
+    > **Note:** Please make sure to select image index in the range of supplied images during application build.
+    By default, pre-built application has 4 images, indexes from 0 to 3.
+
+3. “Run classification on all images” menu option triggers sequential inference executions on all built-in images.
+
+4. “Show NN model info” menu option prints information about model data type, input and output tensor sizes:
+
+    ```log
+    [INFO] uTFL version: 2.5.0
+    [INFO] Model info:
+    [INFO] Model INPUT tensors:
+    [INFO] 	tensor type is UINT8
+    [INFO] 	tensor occupies 150528 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1: 224
+    [INFO] 		2: 224
+    [INFO] 		3:   3
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.007812
+    [INFO] ZeroPoint[0] = 128
+    [INFO] Model OUTPUT tensors:
+    [INFO] 	tensor type is UINT8
+    [INFO] 	tensor occupies 1001 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1: 1001
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.098893
+    [INFO] ZeroPoint[0] = 58
+    [INFO] Activation buffer (a.k.a tensor arena) size used: 521760
+    [INFO] Number of operators: 1
+    [INFO] 	Operator 0: ethos-u
+    [INFO] Use of Arm uNPU is enabled
+    ```
+
+5. “List Images” menu option prints a list of pair image indexes - the original filenames embedded in the application:
+
+    ```log
+    [INFO] List of Files:
+    [INFO] 0 => cat.bmp
+    [INFO] 1 => dog.bmp
+    [INFO] 2 => kimono.bmp
+    [INFO] 3 => tiger.bmp
+    ```
+
+### Running Image Classification
+
+Please select the first menu option to execute Image Classification.
+
+The following example illustrates application output for classification:
+
+```log
+[INFO] Running inference on image 0 => cat.bmp
+[INFO] Profile for Inference:
+	Active NPU cycles: 7622641
+	Idle NPU cycles:   525
+
+[INFO] 0) 282 (14.636096) -> tabby, tabby cat
+[INFO] 1) 286 (14.537203) -> Egyptian cat
+[INFO] 2) 283 (12.757138) -> tiger cat
+[INFO] 3) 458 (7.021370) -> bow tie, bow-tie, bowtie
+[INFO] 4) 288 (7.021370) -> lynx, catamount
+```
+
+It could take several minutes to complete one inference run (average time is 2-3 minutes).
+
+The log shows the inference results for “image 0” (0 - index) that corresponds to “cat.bmp” in the sample image resource
+folder.
+
+The profiling section of the log shows that for this inference:
+
+- Ethos-U55's PMU report:
+
+  - 7,622,641 active cycles: number of NPU cycles that were used for computation
+
+  - 525 idle cycles: number of cycles for which the NPU was idle
+
+- For FPGA platforms, CPU cycle count can also be enabled. For FVP, however, CPU cycle counters should not be used as
+    the CPU model is not cycle-approximate or cycle-accurate.
+
+The application prints the top 5 classes with indexes, confidence score and labels from associated
+labels_mobilenet_v2_1.0_224.txt file. The FVP window also shows the output on its LCD section.
diff --git a/docs/use_cases/inference_runner.md b/docs/use_cases/inference_runner.md
new file mode 100644
index 0000000..ffb205e
--- /dev/null
+++ b/docs/use_cases/inference_runner.md
@@ -0,0 +1,296 @@
+# Inference Runner Code Sample
+
+- [Introduction](#introduction)
+  - [Prerequisites](#prerequisites)
+- [Building the Code Samples application from sources](#building-the-code-samples-application-from-sources)
+  - [Build options](#build-options)
+  - [Build process](#build-process)
+  - [Add custom model](#add-custom-model)
+- [Setting-up and running Ethos-U55 code sample](#setting-up-and-running-ethos-u55-code-sample)
+  - [Setting up the Ethos-U55 Fast Model](#setting-up-the-ethos-u55-fast-model)
+  - [Starting Fast Model simulation](#starting-fast-model-simulation)
+  - [Running Inference Runner](#running-inference-runner)
+- [Inference Runner processing information](#inference-runner-processing-information)
+
+## Introduction
+
+This document describes the process of setting up and running the Arm® Ethos™-U55 NPU Inference Runner.
+The inference runner is intended for quickly checking profiling results for any desired network, providing it has been
+processed by the Vela compiler.
+
+A simple model is provided with the Inference Runner as an example, but it is expected that the user will replace this
+model with one they wish to profile, see [Add custom model](./inference_runner.md#Add-custom-model) for more details.
+
+The inference runner is intended for quickly checking profiling results for any desired network
+providing it has been processed by the Vela compiler.
+
+The inference runner will populate all input tensors for the provided model with randomly generated data and an
+inference is then performed. Profiling results are then displayed in the console.
+
+Use case code could be found in [source/use_case/inference_runner](../../source/use_case/inference_runner]) directory.
+
+### Prerequisites
+
+See [Prerequisites](../documentation.md#prerequisites)
+
+## Building the Code Samples application from sources
+
+### Build options
+
+In addition to the already specified build option in the main documentation, the Inference Runner use case adds:
+
+- `inference_runner_MODEL_TFLITE_PATH` - Path to the NN model file in TFLite format. Model will be processed and
+  included into the application axf file. The default value points to one of the delivered set of models.
+  Note that the parameters `TARGET_PLATFORM` and `ETHOS_U55_ENABLED` should be aligned with the chosen model, i.e.:
+  - if `ETHOS_U55_ENABLED` is set to `On` or `1`, the NN model is assumed to be optimized. The model will naturally
+    all back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
+  - if `ETHOS_U55_ENABLED` is set to `Off` or `0`, the NN model is assumed to be unoptimized. Supplying an optimized model
+    in this case will result in a runtime error.
+
+- `inference_runner_ACTIVATION_BUF_SZ`: The intermediate/activation buffer size reserved for the NN model. By
+    default, it is set to 2MiB and should be enough for most models.
+
+In order to build **ONLY** Inference Runner example application add to the `cmake` command line specified in
+[Building](../documentation.md#Building) `-DUSE_CASE_BUILD=inferece_runner`.
+
+### Build process
+
+> **Note:** This section describes the process for configuring the build for `MPS3: SSE-300` for different target platform
+>see [Building](../documentation.md#Building) section.
+
+Create a build directory and navigate inside:
+
+```commandline
+mkdir build_inference_runner && cd build_inference_runner
+```
+
+On Linux, execute the following command to build **only** Inference Runner application to run on the Ethos-U55 Fast
+Model when providing only the mandatory arguments for CMake configuration:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=inference_runner ..
+```
+
+For Windows, add `-G "MinGW Makefiles"`:
+
+```commandline
+cmake \
+    -G "MinGW Makefiles" \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=inference_runner ..
+```
+
+Toolchain option `CMAKE_TOOLCHAIN_FILE` points to the toolchain specific file to set the compiler and platform specific
+parameters.
+
+To configure a build that can be debugged using Arm-DS, we can just specify
+the build type as `Debug`:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DUSE_CASE_BUILD=inference_runner ..
+```
+
+To configure a build that can be debugged using a tool that only supports
+DWARF format 3 (Modeldebugger for example), we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DARMCLANG_DEBUG_DWARF_LEVEL=3 \
+    -DUSE_CASE_BUILD=inference_runner ..
+```
+
+> **Note:** If building for different Ethos-U55 configurations, see
+>[Configuring build for different Arm Ethos-U55 configurations](../sections/building.md#Configuring-build-for-different-Arm-Ethos-U55-configurations):
+
+If the TensorFlow source tree is not in its default expected location,
+set the path using `TENSORFLOW_SRC_PATH`.
+Similarly, if the Ethos-U55 driver is not in the default location,
+`ETHOS_U55_DRIVER_SRC_PATH` can be used to configure the location. For example:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=/my/custom/location/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=/my/custom/location/core_driver \
+    -DUSE_CASE_BUILD=inference_runner ..
+```
+
+Also, `CMSIS_SRC_PATH` parameter can be used to override the CMSIS sources used for compilation used by TensorFlow by
+default. For example, to use the CMSIS sources fetched by the ethos-u helper script, we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=../ethos-u/core_software/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=../ethos-u/core_software/core_driver \
+    -DCMSIS_SRC_PATH=../ethos-u/core_software/cmsis \
+    -DUSE_CASE_BUILD=inference_runner ..
+```
+
+> **Note:** If re-building with changed parameters values, it is highly advised to clean the build directory and re-run
+>the CMake command.
+
+If the CMake command succeeded, build the application as follows:
+
+```commandline
+make -j4
+```
+
+For Windows, use `mingw32-make`.
+
+Add VERBOSE=1 to see compilation and link details.
+
+Results of the build will be placed under `build/bin` folder:
+
+```tree
+bin
+ ├── ethos-u-inference_runner.axf
+ ├── ethos-u-inference_runner.htm
+ ├── ethos-u-inference_runner.map
+ ├── images-inference_runner.txt
+ └── sectors
+      ├── kws
+      │ └── ...
+      └── img_class
+        ├── dram.bin
+        └── itcm.bin
+```
+
+Where:
+
+- `ethos-u-inference_runner.axf`: The built application binary for the Inference Runner use case.
+
+- `ethos-u-inference_runner.map`: Information from building the application (e.g. libraries used, what was optimized,
+    location of objects)
+
+- `ethos-u-inference_runner.htm`: Human readable file containing the call graph of application functions.
+
+- `sectors/`: Folder containing the built application, split into files for loading into different FPGA memory regions.
+
+- `Images-inference_runner.txt`: Tells the FPGA which memory regions to use for loading the binaries in sectors/**
+    folder.
+
+### Add custom model
+
+The application performs inference using the model pointed to by the CMake parameter `inference_runner_MODEL_TFLITE_PATH`.
+
+> **Note:** If you want to run the model using Ethos-U55, ensure your custom model has been run through the Vela compiler
+>successfully before continuing. See [Optimize model with Vela compiler](../sections/building.md#Optimize-custom-model-with-Vela-compiler).
+
+Then, you must set `inference_runner_MODEL_TFLITE_PATH` to the location of the Vela processed model file.
+
+An example:
+
+```commandline
+cmake \
+  -Dinference_runner_MODEL_TFLITE_PATH=<path/to/custom_model_after_vela.tflite> \
+  -DTARGET_PLATFORM=mps3 \
+  -DTARGET_SUBSYSTEM=sse-300 \
+  -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+The `.tflite` model file pointed to by `inference_runner_MODEL_TFLITE_PATH` will be converted to C++ files during the CMake
+configuration stage and then compiled into the application for performing inference with.
+
+The log from the configuration stage should tell you what model path has been used:
+
+```stdout
+-- User option inference_runner_MODEL_TFLITE_PATH is set to <path/to/custom_model_after_vela.tflite>
+...
+-- Using <path/to/custom_model_after_vela.tflite>
+++ Converting custom_model_after_vela.tflite to\
+custom_model_after_vela.tflite.cc
+...
+```
+
+After compiling, your custom model will have now replaced the default one in the application.
+
+## Setting-up and running Ethos-U55 code sample
+
+### Setting up the Ethos-U55 Fast Model
+
+The FVP is available publicly from
+[Arm Ecosystem FVP downloads](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps).
+
+For Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains Ethos-U55 and
+Cortex-M55. The model is currently only supported on Linux based machines. To install the FVP:
+
+- Unpack the archive
+
+- Run the install script in the extracted package
+
+```commandline
+./FVP_Corstone_SSE-300_Ethos-U55.sh
+```
+
+- Follow the instructions to install the FVP to your desired location
+
+### Starting Fast Model simulation
+
+Once completed the building step, application binary ethos-u-infernce_runner.axf can be found in the `build/bin` folder.
+Assuming the install location of the FVP was set to ~/FVP_install_location, the simulation can be started by:
+
+```commandline
+~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55
+./bin/mps3-sse-300/ethos-u-inference_runner.axf
+```
+
+A log output should appear on the terminal:
+
+```log
+telnetterminal0: Listening for serial connection on port 5000
+telnetterminal1: Listening for serial connection on port 5001
+telnetterminal2: Listening for serial connection on port 5002
+telnetterminal5: Listening for serial connection on port 5003
+```
+
+This will also launch a telnet window with the sample application's standard output and error log entries containing
+information about the pre-built application version, TensorFlow Lite Micro library version used, data type as well as
+the input and output tensor sizes of the model compiled into the executable binary.
+
+### Running Inference Runner
+
+After the application has started the inference starts immediately and it outputs the results on the telnet terminal.
+
+The following example illustrates application output:
+
+```log
+[INFO] Profile for Inference:
+       Active NPU cycles: 26976
+       Idle NPU cycles: 196
+```
+
+After running an inference on randomly generated data, the output of the log shows the profiling results that for this
+inference:
+
+- Ethos-U55's PMU report:
+
+  - 26,976 active cycles: number of cycles that were used for computation
+
+  - 196 idle cycles: number of cycles for which the NPU was idle
+
+- For FPGA platforms, CPU cycle count can also be enabled. For FVP, however, CPU cycle counters should not be used as
+    the CPU model is not cycle-approximate or cycle-accurate.
diff --git a/docs/use_cases/kws.md b/docs/use_cases/kws.md
new file mode 100644
index 0000000..316b501
--- /dev/null
+++ b/docs/use_cases/kws.md
@@ -0,0 +1,474 @@
+# Keyword Spotting Code Sample
+
+- [Introduction](#introduction)
+  - [Prerequisites](#prerequisites)
+- [Building the code sample application from sources](#building-the-code-sample-application-from-sources)
+  - [Build options](#build-options)
+  - [Build process](#build-process)
+  - [Add custom input](#add-custom-input)
+  - [Add custom model](#add-custom-model)
+- [Setting-up and running Ethos-U55 code sample](#setting-up-and-running-ethos-u55-code-sample)
+  - [Setting up the Ethos-U55 Fast Model](#setting-up-the-ethos-u55-fast-model)
+  - [Starting Fast Model simulation](#starting-fast-model-simulation)
+  - [Running Keyword Spotting](#running-keyword-spotting)
+- [Keyword Spotting processing information](#keyword-spotting-processing-information)
+  - [Preprocessing and feature extraction](#preprocessing-and-feature-extraction)
+  - [Postprocessing](#postprocessing)
+
+## Introduction
+
+This document describes the process of setting up and running the Arm® Ethos™-U55 Keyword Spotting
+example.
+
+Use case code could be found in [source/use_case/kws](../../source/use_case/kws]) directory.
+
+### Preprocessing and feature extraction
+
+The DS-CNN keyword spotting model that is supplied with the Code Samples expects audio data to be preprocessed in
+a specific way before performing an inference. This section aims to provide an overview of the feature extraction
+process used.
+
+First the audio data is normalized to the range (-1, 1).
+
+> **Note:** Mel-frequency cepstral coefficients (MFCCs) are a common feature extracted from audio data and can be used as
+>input for machine learning tasks like keyword spotting and speech recognition.
+>See source/application/main/include/Mfcc.hpp for implementation details.
+
+Next, a window of 640 audio samples is taken from the start of the audio clip. From these 640 samples we calculate 10
+MFCC features.
+
+The whole window is shifted to the right by 320 audio samples and 10 new MFCC features are calculated. This process of
+shifting and calculating is repeated until the end of the 16000 audio samples needed to perform an inference is reached.
+In total this will be 49 windows that each have 10 MFCC features calculated for them, giving an input tensor of shape
+49x10.
+
+These extracted features are quantized, and an inference is performed.
+
+![KWS preprocessing](../media/KWS_preprocessing.png)
+
+If the audio clip is longer than 16000 audio samples then the initial starting position is offset by 16000/2 = 8000
+audio samples. From this new starting point, MFCC features for the next 16000 audio samples are calculated and another
+inference is performed (i.e. do an inference for samples 8000-24000).
+
+> **Note:** Parameters of the MFCC feature extraction such as window size, stride, number of features etc. all depend on
+>what was used during model training. These values are specific to each model and if you try a different keyword spotting
+>model that uses MFCC input then values are likely to need changing to match the new model.
+In addition, MFCC feature extraction methods can vary slightly with different normalization methods or scaling etc. being used.
+
+### Postprocessing
+
+After an inference is complete the highest probability detected word is output to console, providing its probability is
+larger than a threshold value (default 0.9).
+
+If multiple inferences are performed for an audio clip, then multiple results will be output.
+
+### Prerequisites
+
+See [Prerequisites](../documentation.md#prerequisites)
+
+## Building the code sample application from sources
+
+### Build options
+
+In addition to the already specified build option in the main documentation, keyword spotting use case adds:
+
+- `kws_MODEL_TFLITE_PATH` - Path to the NN model file in TFLite format. Model will be processed and included into the application axf file. The default value points to one of the delivered set of models. Note that the parameters `kws_LABELS_TXT_FILE`,`TARGET_PLATFORM` and `ETHOS_U55_ENABLED` should be aligned with the chosen model, i.e.:
+  - if `ETHOS_U55_ENABLED` is set to `On` or `1`, the NN model is assumed to be optimized. The model will naturally fall back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
+  - if `ETHOS_U55_ENABLED` is set to `Off` or `0`, the NN model is assumed to be unoptimized. Supplying an optimized model in this case will result in a runtime error.
+
+- `kws_FILE_PATH`: Path to the directory containing audio files, or a path to single WAV file, to be used in the application. The default value points
+    to the resources/kws/samples folder containing the delivered set of audio clips.
+
+- `kws_LABELS_TXT_FILE`: Path to the labels' text file. The file is used to map key word class index to the text
+    label. The default value points to the delivered labels.txt file inside the delivery package.
+
+- `kws_AUDIO_RATE`: Input data sampling rate. Each audio file from kws_FILE_PATH is preprocessed during the build to
+    match NN model input requirements. Default value is 16000.
+
+- `kws_AUDIO_MONO`: If set to ON the audio data will be converted to mono. Default is ON.
+
+- `kws_AUDIO_OFFSET`: Start loading audio data starting from this offset (in seconds). Default value is 0.
+
+- `kws_AUDIO_DURATION`: Length of the audio data to be used in the application in seconds. Default is 0 meaning the
+    whole audio file will be taken.
+
+- `kws_AUDIO_MIN_SAMPLES`: Minimum number of samples required by the network model. If the audio clip is shorter than
+    this number, it is padded with zeros. Default value is 16000.
+
+- `kws_MODEL_SCORE_THRESHOLD`: Threshold value [0.0, 1.0] that must be applied to the inference results for a
+    label to be deemed valid. Default is 0.9
+
+- `kws_ACTIVATION_BUF_SZ`: The intermediate/activation buffer size reserved for the NN model. By default, it is set
+    to 1MiB and should be enough for most models.
+
+In order to build **ONLY** keyword spotting example application add to the `cmake` command line specified in [Building](../documentation.md#Building) `-DUSE_CASE_BUILD=kws`.
+
+### Build process
+
+> **Note:** This section describes the process for configuring the build for `MPS3: SSE-300` for different target platform see [Building](../documentation.md#Building) section.
+
+In order to build **only** the keyword spotting example, create a build directory and
+navigate inside, for example:
+
+```commandline
+mkdir build_kws && cd build_kws
+```
+
+On Linux, execute the following command to build Keyword Spotting application to run on the Ethos-U55 Fast Model when providing only the mandatory arguments for CMake configuration:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=kws ..
+```
+
+For Windows, add `-G "MinGW Makefiles"`:
+
+```commandline
+cmake \
+    -G "MinGW Makefiles" \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=kws ..
+```
+
+Toolchain option `CMAKE_TOOLCHAIN_FILE` points to the toolchain specific file to set the compiler and platform specific
+parameters.
+
+To configure a build that can be debugged using Arm-DS, we can just specify
+the build type as `Debug`:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DUSE_CASE_BUILD=kws ..
+```
+
+To configure a build that can be debugged using a tool that only supports
+DWARF format 3 (Modeldebugger for example), we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DARMCLANG_DEBUG_DWARF_LEVEL=3 \
+    -DUSE_CASE_BUILD=kws ..
+```
+
+> **Note:** If building for different Ethos-U55 configurations, see [Configuring build for different Arm Ethos-U55 configurations](../sections/building.md#Configuring-build-for-different-Arm-Ethos-U55-configurations):
+
+If the TensorFlow source tree is not in its default expected location,
+set the path using `TENSORFLOW_SRC_PATH`.
+Similarly, if the Ethos-U55 driver is not in the default location,
+`ETHOS_U55_DRIVER_SRC_PATH` can be used to configure the location. For example:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=/my/custom/location/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=/my/custom/location/core_driver \
+    -DUSE_CASE_BUILD=kws ..
+```
+
+Also, `CMSIS_SRC_PATH` parameter can be used to override the CMSIS sources used for compilation used by TensorFlow by default. For example, to use the CMSIS sources fetched by the ethos-u helper script, we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=../ethos-u/core_software/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=../ethos-u/core_software/core_driver \
+    -DCMSIS_SRC_PATH=../ethos-u/core_software/cmsis \
+    -DUSE_CASE_BUILD=kws ..
+```
+
+> **Note:** If re-building with changed parameters values, it is highly advised to clean the build directory and re-run the CMake command.
+
+If the CMake command succeeded, build the application as follows:
+
+```commandline
+make -j4
+```
+
+For Windows, use `mingw32-make`.
+
+Add VERBOSE=1 to see compilation and link details.
+
+Results of the build will be placed under `build/bin` folder:
+
+```tree
+bin
+ ├── ethos-u-kws.axf
+ ├── ethos-u-kws.htm
+ ├── ethos-u-kws.map
+ ├── images-kws.txt
+ └── sectors
+      └── kws
+           ├── dram.bin
+           └── itcm.bin
+```
+
+Where:
+
+- `ethos-u-kws.axf`: The built application binary for the Keyword Spotting use case.
+
+- `ethos-u-kws.map`: Information from building the application (e.g. libraries used, what was optimized, location of
+    objects)
+
+- `ethos-u-kws.htm`: Human readable file containing the call graph of application functions.
+
+- `sectors/`: Folder containing the built application, split into files for loading into different FPGA memory regions.
+
+- `Images-kws.txt`: Tells the FPGA which memory regions to use for loading the binaries in sectors/\*\* folder.
+
+### Add custom input
+
+The application performs inference on audio data found in the folder, or an individual file, set by the CMake parameter `kws_FILE_PATH`.
+
+To run the application with your own audio clips first create a folder to hold them and then copy the custom audio files
+into this folder, for example:
+
+```commandline
+mkdir /tmp/custom_wavs
+
+cp my_clip.wav /tmp/custom_wavs/
+```
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+Next set `kws_FILE_PATH` to the location of this folder when building:
+
+```commandline
+cmake \
+    -Dkws_FILE_PATH=/tmp/custom_wavs/ \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DUSE_CASE_BUILD=kws \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+The audio clips found in the `kws_FILE_PATH` folder will be picked up and automatically converted to C++ files during the
+CMake configuration stage and then compiled into the application during the build phase for performing inference with.
+
+The log from the configuration stage should tell you what audio clip directory path has been used:
+
+```log
+-- User option kws_FILE_PATH is set to /tmp/custom_wavs
+-- Generating audio files from /tmp/custom_wavs
+++ Converting my_clip.wav to my_clip.cc
+++ Generating build/generated/kws/include/AudioClips.hpp
+++ Generating build/generated/kws/src/AudioClips.cc
+-- Defined build user options:
+-- kws_FILE_PATH=/tmp/custom_wavs
+```
+
+After compiling, your custom inputs will have now replaced the default ones in the application.
+
+> **Note:** The CMake parameter `kws_AUDIO_MIN_SAMPLES` determine the minimum number of input sample. When building the application,
+if the size of the audio clips is less then `kws_AUDIO_MIN_SAMPLES` then it will be padded so that it does.
+
+### Add custom model
+
+The application performs inference using the model pointed to by the CMake parameter `kws_MODEL_TFLITE_PATH`.
+
+> **Note:** If you want to run the model using Ethos-U55, ensure your custom model has been run through the Vela compiler successfully before continuing. See [Optimize model with Vela compiler](../sections/building.md#Optimize-custom-model-with-Vela-compiler).
+
+To run the application with a custom model you will need to provide a labels_<model_name>.txt file of labels
+associated with the model. Each line of the file should correspond to one of the outputs in your model. See the provided
+ds_cnn_labels.txt file for an example.
+
+Then, you must set kws_MODEL_TFLITE_PATH to the location of the Vela processed model file and kws_LABELS_TXT_FILE
+to the location of the associated labels file.
+
+An example:
+
+```commandline
+cmake \
+    -Dkws_MODEL_TFLITE_PATH=<path/to/custom_model_after_vela.tflite> \
+    -Dkws_LABELS_TXT_FILE=<path/to/labels_custom_model.txt> \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DUSE_CASE_BUILD=kws \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+The `.tflite` model file pointed to by `kws_MODEL_TFLITE_PATH` and labels text file pointed to by `kws_LABELS_TXT_FILE` will
+be converted to C++ files during the CMake configuration stage and then compiled into the application for performing
+inference with.
+
+The log from the configuration stage should tell you what model path and labels file have been used:
+
+```log
+-- User option kws_MODEL_TFLITE_PATH is set to <path/to/custom_model_after_vela.tflite>
+...
+-- User option kws_LABELS_TXT_FILE is set to <path/to/labels_custom_model.txt>
+...
+-- Using <path/to/custom_model_after_vela.tflite>
+++ Converting custom_model_after_vela.tflite to\
+custom_model_after_vela.tflite.cc
+-- Generating labels file from <path/to/labels_custom_model.txt>
+-- writing to <path/to/build/generated/src/Labels.cc>
+...
+```
+
+After compiling, your custom model will have now replaced the default one in the application.
+
+## Setting-up and running Ethos-U55 code sample
+
+### Setting up the Ethos-U55 Fast Model
+
+The FVP is available publicly from [Arm Ecosystem FVP downloads](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps).
+
+For Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains Ethos-U55 and
+Cortex-M55. The model is currently only supported on Linux based machines. To install the FVP:
+
+- Unpack the archive
+
+- Run the install script in the extracted package
+
+```commandline
+./FVP_Corstone_SSE-300_Ethos-U55.sh
+```
+
+- Follow the instructions to install the FVP to your desired location
+
+### Starting Fast Model simulation
+
+Once completed the building step, application binary ethos-u-kws.axf can be found in the `build/bin` folder.
+Assuming the install location of the FVP was set to ~/FVP_install_location, the simulation can be started by:
+
+```commandline
+~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55
+./bin/mps3-sse-300/ethos-u-kws.axf
+```
+
+A log output should appear on the terminal:
+
+```log
+telnetterminal0: Listening for serial connection on port 5000
+telnetterminal1: Listening for serial connection on port 5001
+telnetterminal2: Listening for serial connection on port 5002
+telnetterminal5: Listening for serial connection on port 5003
+```
+
+This will also launch a telnet window with the sample application's standard output and error log entries containing
+information about the pre-built application version, TensorFlow Lite Micro library version used, data type as well as
+the input and output tensor sizes of the model compiled into the executable binary.
+
+After the application has started if `kws_FILE_PATH` pointed to a single file (or a folder containing a single input file)
+the inference starts immediately. In case of multiple inputs choice, it outputs a menu and waits for the user input from telnet terminal:
+
+```log
+User input required
+Enter option number from:
+
+1. Classify next audio clip
+2. Classify audio clip at chosen index
+3. Run classification on all audio clips
+4. Show NN model info
+5. List audio clips
+
+Choice:
+
+```
+
+1. “Classify next audio clip” menu option will run inference on the next in line voice clip from the collection of the
+    compiled audio.
+
+    > **Note:** Note that if the clip is over a certain length, the application will invoke multiple inference runs to cover the entire file.
+
+2. “Classify audio clip at chosen index” menu option will run inference on the chosen audio clip.
+
+    > **Note:** Please make sure to select audio clip index in the range of supplied audio clips during application build.
+    By default, pre-built application has 4 files, indexes from 0 to 3.
+
+3. “Run classification on all audio clips” menu option triggers sequential inference executions on all built-in voice
+    samples.
+
+4. “Show NN model info” menu option prints information about model data type, input and output tensor sizes:
+
+    ```log
+    [INFO] uTFL version: 2.5.0
+    [INFO] Model info:
+    [INFO] Model INPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 490 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:   1
+    [INFO] 		2:  49
+    [INFO] 		3:  10
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 1.107164
+    [INFO] ZeroPoint[0] = 95
+    [INFO] Model OUTPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 12 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:  12
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.003906
+    [INFO] ZeroPoint[0] = -128
+    [INFO] Activation buffer (a.k.a tensor arena) size used: 72848
+    [INFO] Number of operators: 1
+    [INFO] 	Operator 0: ethos-u
+    [INFO] Use of Arm uNPU is enabled
+    ```
+
+5. “List audio clips” menu option prints a list of pair audio indexes - the original filenames embedded in the
+    application:
+
+    ```log
+    [INFO] List of Files:
+    [INFO] 0 => down.wav
+    [INFO] 1 => rightleftup.wav
+    [INFO] 2 => yes.wav
+    [INFO] 3 => yesnogostop.wav
+    ```
+
+### Running Keyword Spotting
+
+Selecting the first option will run inference on the first file.
+
+The following example illustrates application output for classification:
+
+```log
+[INFO] Running inference on audio clip 0 => down.wav
+[INFO] Inference 1/1
+[INFO] Profile for Inference:
+	Active NPU cycles: 680400
+	Idle NPU cycles:   766
+
+[INFO] For timestamp: 0.000000 (inference #: 0); threshold: 0.900000
+[INFO] 		label @ 0: down, score: 0.996094
+```
+
+Each inference should take less than 30 seconds on most systems running Fast Model.
+The profiling section of the log shows that for this inference:
+
+- Ethos-U55's PMU report:
+
+  - 680,400 active cycles: number of cycles that were used for computation
+
+  - 766 idle cycles: number of cycles for which the NPU was idle
+
+- For FPGA platforms, CPU cycle count can also be enabled. For FVP, however, CPU cycle counters should not be used as
+    the CPU model is not cycle-approximate or cycle-accurate.
+
+The application prints the highest confidence score and the associated label from ds_cnn_labels.txt file.
\ No newline at end of file
diff --git a/docs/use_cases/kws_asr.md b/docs/use_cases/kws_asr.md
new file mode 100644
index 0000000..e79b887
--- /dev/null
+++ b/docs/use_cases/kws_asr.md
@@ -0,0 +1,589 @@
+# Keyword Spotting and Automatic Speech Recognition Code Sample
+
+- [Introduction](#introduction)
+  - [Prerequisites](#prerequisites)
+- [Building the code sample application from sources](#building-the-code-sample-application-from-sources)
+  - [Build options](#build-options)
+  - [Build process](#build-process)
+  - [Add custom input](#add-custom-input)
+  - [Add custom model](#add-custom-model)
+- [Setting-up and running Ethos-U55 Code Samples](#setting-up-and-running-ethos-u55-code-samples)
+  - [Setting up the Ethos-U55 Fast Model](#setting-up-the-ethos-u55-fast-model)
+  - [Starting Fast Model simulation](#starting-fast-model-simulation)
+  - [Running Keyword Spotting and Automatic Speech Recognition](#running-keyword-spotting-and-automatic-speech-recognition)
+- [Keyword Spotting and Automatic Speech Recognition processing information](#keyword-spotting-and-automatic-speech-recognition-processing-information)
+  - [Preprocessing and feature extraction](#preprocessing-and-feature-extraction)
+    - [Keyword Spotting Preprocessing](#keyword-spotting-preprocessing)
+    - [Automatic Speech Recognition Preprocessing](#automatic-speech-recognition-preprocessing)
+  - [Postprocessing](#postprocessing)
+
+## Introduction
+
+This document describes the process of setting up and running an example of sequential execution of the Keyword Spotting
+and Automatic Speech Recognition models on Cortex-M CPU and Ethos-U NPU.
+
+The Keyword Spotting and Automatic Speech Recognition example demonstrates how to run multiple models sequentially. A
+Keyword Spotting model is first run on the CPU and if a set keyword is detected then an Automatic Speech Recognition
+model is run on Ethos-U55 on the remaining audio.
+Tensor arena memory region is reused between models to optimise application memory footprint.
+
+"Yes" key word is used to trigger full command recognition following the key word.
+Use case code could be found in [source/use_case/kws_asr](../../source/use_case/kws_asr]) directory.
+
+### Preprocessing and feature extraction
+
+In this use-case there are 2 different models being used with different requirements for preprocessing. As such each
+preprocessing process is detailed below. Note that Automatic Speech Recognition only occurs if a keyword is detected in
+the audio clip.
+
+By default the KWS model is run purely on CPU and not on the Ethos-U55.
+
+#### Keyword Spotting Preprocessing
+
+The DS-CNN keyword spotting model that is used with the Code Samples expects audio data to be preprocessed in
+a specific way before performing an inference. This section aims to provide an overview of the feature extraction
+process used.
+
+First the audio data is normalized to the range (-1, 1).
+
+> **Note:** Mel-frequency cepstral coefficients (MFCCs) are a common feature extracted from audio data and can be used as input for machine learning tasks like keyword spotting and speech recognition. See source/application/main/include/Mfcc.hpp for implementation details.
+
+Next, a window of 640 audio samples is taken from the start of the audio clip. From these 640 samples we calculate 10
+MFCC features.
+
+The whole window is shifted to the right by 320 audio samples and 10 new MFCC features are calculated. This process of
+shifting and calculating is repeated until the end of the 16000 audio samples needed to perform an inference is reached.
+In total this will be 49 windows that each have 10 MFCC features calculated for them, giving an input tensor of shape
+49x10.
+
+These extracted features are quantized, and an inference is performed.
+
+If the audio clip is longer than 16000 audio samples then the initial starting position is offset by 16000/2 = 8000
+audio samples. From this new starting point, MFCC features for the next 16000 audio samples are calculated and another
+inference is performed (i.e. do an inference for samples 8000-24000).
+
+> **Note:** Parameters of the MFCC feature extraction such as window size, stride, number of features etc. all depend on what was used during model training. These values are specific to each model and if you try a different keyword spotting model that uses MFCC input then values are likely to need changing to match the new model.
+
+In addition, MFCC feature extraction methods can vary slightly with different normalization methods or scaling etc. being used.
+
+#### Automatic Speech Recognition Preprocessing
+
+The wav2letter automatic speech recognition model that is used with the Code Samples expects audio data to be
+preprocessed in a specific way before performing an inference. This section aims to provide an overview of the feature
+extraction process used.
+
+First the audio data is normalized to the range (-1, 1).
+
+> **Note:** Mel-frequency cepstral coefficients (MFCCs) are a common feature extracted from audio data and can be used as input for machine learning tasks like keyword spotting and speech recognition. See source/application/main/include/Mfcc.hpp for implementation details.
+
+Next, a window of 512 audio samples is taken from the start of the audio clip. From these 512 samples we calculate 13
+MFCC features.
+
+The whole window is shifted to the right by 160 audio samples and 13 new MFCC features are calculated. This process of
+shifting and calculating is repeated until enough audio samples to perform an inference have been processed. In total
+this will be 296 windows that each have 13 MFCC features calculated for them.
+
+After extracting MFCC features the first and second order derivatives of these features with respect to time are
+calculated. These derivative features are then standardized and concatenated with the MFCC features (which also get
+standardized). At this point the input tensor will have a shape of 296x39.
+
+These extracted features are quantized, and an inference is performed.
+
+For longer audio clips where multiple inferences need to be performed, then the initial starting position is offset by
+(100\*160) = 16000 audio samples. From this new starting point, MFCC and derivative features are calculated as before
+until there is enough to perform another inference. Padding can be used if there are not enough audio samples for at
+least 1 inference. This step is repeated until the whole audio clip has been processed. If there are not enough audio
+samples for a final complete inference the MFCC features will be padded by repeating the last calculated feature until
+an inference can be performed.
+
+> **Note:** Parameters of the MFCC feature extraction such as window size, stride, number of features etc. all depend on what was used during model training. These values are specific to each model. If you switch to a different ASR model than the one supplied, then the feature extraction process could be completely different to the one currently implemented.
+
+The amount of audio samples we offset by for long audio clips is specific to the included wav2letter model.
+
+### Postprocessing
+
+If a keyword is detected then the ASR process is run and the raw output of that inference needs to be postprocessed to
+get a usable result.
+
+The raw output from the model is a tensor of shape 148x29 where each row is a probability distribution over the possible
+29 characters that can appear at each of the 148 time steps.
+
+This wav2letter model is trained using context windows, this means that only certain parts of the output are usable
+depending on the bit of the audio clip that is currently being processed.
+
+If this is the first inference and multiple inferences are required, then ignore the final 49 rows of the output.
+Similarly, if this is the final inference from multiple inferences then ignore the first 49 rows of the output. Finally,
+if this inference is not the last or first inference then ignore the first and last 49 rows of the model output.
+
+> **Note:** If the audio clip is small enough then the whole of the model output is usable and there is no need to throw away any of the output before continuing.
+
+Once any rows have been removed the final processing can be done. To process the output, first the letter with the
+highest probability at each time step is found. Next, any letters that are repeated multiple times in a row are removed
+(e.g. [t, t, t, o, p, p] becomes [t, o, p]). Finally, the 29^th^ blank token letter is removed from the output.
+
+For the final output, the result from all inferences are combined before decoding. What you are left with is then
+displayed to the console.
+
+### Prerequisites
+
+See [Prerequisites](../documentation.md#prerequisites)
+
+## Building the code sample application from sources
+
+### Build options
+
+In addition to the already specified build option in the main documentation, Keyword Spotting and Automatic Speech
+Recognition use case adds:
+
+- `kws_asr_MODEL_TFLITE_PATH_ASR` and `kws_asr_MODEL_TFLITE_PATH_KWS`: Path to the NN model files in TFLite format.
+    Models will be processed and included into the application axf file. The default value points to one of the delivered set of models.
+    Note that the parameters `kws_asr_LABELS_TXT_FILE_KWS`, `kws_asr_LABELS_TXT_FILE_ASR`,`TARGET_PLATFORM` and `ETHOS_U55_ENABLED`
+    should be aligned with the chosen model, i.e:
+  - if `ETHOS_U55_ENABLED` is set to `On` or `1`, the NN model is assumed to be optimized. The model will naturally fall back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
+  - if `ETHOS_U55_ENABLED` is set to `Off` or `0`, the NN model is assumed to be unoptimized. Supplying an optimized model in this case will result in a runtime error.
+
+- `kws_asr_FILE_PATH`: Path to the directory containing audio files, or a path to single WAV file, to be used in the application. The default value
+    points to the resources/kws_asr/samples folder containing the delivered set of audio clips.
+
+- `kws_asr_LABELS_TXT_FILE_KWS` and `kws_asr_LABELS_TXT_FILE_ASR`: Path respectively to keyword spotting labels' and the automatic speech
+    recognition labels' text files. The file is used to map
+    letter class index to the text label. The default value points to the delivered labels.txt file inside the delivery
+    package.
+
+- `kws_asr_AUDIO_RATE`: Input data sampling rate. Each audio file from kws_asr_FILE_PATH is preprocessed during the
+    build to match NN model input requirements. Default value is 16000.
+
+- `kws_asr_AUDIO_MONO`: If set to ON the audio data will be converted to mono. Default is ON.
+
+- `kws_asr_AUDIO_OFFSET`: Start loading audio data starting from this offset (in seconds). Default value is 0.
+
+- `kws_asr_AUDIO_DURATION`: Length of the audio data to be used in the application in seconds. Default is 0 meaning
+    the whole audio file will be taken.
+
+- `kws_asr_AUDIO_MIN_SAMPLES`: Minimum number of samples required by the network model. If the audio clip is shorter
+    than this number, it is padded with zeros. Default value is 16000.
+
+- `kws_asr_MODEL_SCORE_THRESHOLD_KWS`: Threshold value that must be applied to the keyword spotting inference
+    results for a label to be deemed valid. Default is 0.9.
+
+- `kws_asr_MODEL_SCORE_THRESHOLD_ASR`: Threshold value that must be applied to the automatic speech recognition
+    inference results for a label to be deemed valid. Default is 0.5.
+
+- `kws_asr_ACTIVATION_BUF_SZ`: The intermediate/activation buffer size reserved for the NN model. By default, it is
+    set to 2MiB and should be enough for most models.
+
+In order to build **ONLY** Keyword Spotting and Automatic Speech
+Recognition example application add to the `cmake` command line specified in [Building](../documentation.md#Building) `-DUSE_CASE_BUILD=kws_asr`.
+
+### Build process
+
+> **Note:** This section describes the process for configuring the build for `MPS3: SSE-300` for different target platform see [Building](../documentation.md#Building).
+
+Create a build directory and navigate inside:
+
+```commandline
+mkdir build_kws_asr && cd build_kws_asr
+```
+
+On Linux, execute the following command to build the application to run on the Ethos-U55 Fast Model when providing only the mandatory arguments for CMake configuration:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+For Windows, add `-G "MinGW Makefiles"`:
+
+```commandline
+cmake \
+    -G "MinGW Makefiles" \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=./scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+Toolchain option `CMAKE_TOOLCHAIN_FILE` points to the toolchain specific file to set the compiler and platform specific
+parameters.
+
+To configure a build that can be debugged using Arm-DS, we can just specify
+the build type as `Debug`:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+To configure a build that can be debugged using a tool that only supports
+DWARF format 3 (Modeldebugger for example), we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DCMAKE_BUILD_TYPE=Debug \
+    -DARMCLANG_DEBUG_DWARF_LEVEL=3 \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+> **Note:** If building for different Ethos-U55 configurations, see [Configuring build for different Arm Ethos-U55 configurations](../sections/building.md#Configuring-build-for-different-Arm-Ethos-U55-configurations):
+
+If the TensorFlow source tree is not in its default expected location,
+set the path using `TENSORFLOW_SRC_PATH`.
+Similarly, if the Ethos-U55 driver is not in the default location,
+`ETHOS_U55_DRIVER_SRC_PATH` can be used to configure the location. For example:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=/my/custom/location/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=/my/custom/location/core_driver \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+Also, `CMSIS_SRC_PATH` parameter can be used to override the CMSIS sources used for compilation used by TensorFlow by default. For example, to use the CMSIS sources fetched by the ethos-u helper script, we can use:
+
+```commandline
+cmake \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DTENSORFLOW_SRC_PATH=../ethos-u/core_software/tensorflow \
+    -DETHOS_U55_DRIVER_SRC_PATH=../ethos-u/core_software/core_driver \
+    -DCMSIS_SRC_PATH=../ethos-u/core_software/cmsis \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+> **Note:** If re-building with changed parameters values, it is highly advised to clean the build directory and re-run the CMake command.
+
+If the CMake command succeeded, build the application as follows:
+
+```commandline
+make -j4
+```
+
+For Windows, use `mingw32-make`.
+
+Add VERBOSE=1 to see compilation and link details.
+
+Results of the build will be placed under `build/bin` folder:
+
+```tree
+bin
+ ├── ethos-u-kws_asr.axf
+ ├── ethos-u-kws_asr.htm
+ ├── ethos-u-kws_asr.map
+ ├── images-kws_asr.txt
+ └── sectors
+      └── kws_asr
+           ├── dram.bin
+           └── itcm.bin
+```
+
+Where:
+
+- `ethos-u-kws_asr.axf`: The built application binary for the Keyword Spotting and Automatic Speech Recognition use
+    case.
+
+- `ethos-u-kws_asr.map`: Information from building the application (e.g. libraries used, what was optimized, location
+    of objects)
+
+- `ethos-u-kws_asr.htm`: Human readable file containing the call graph of application functions.
+
+- `sectors/`: Folder containing the built application, split into files for loading into different FPGA memory regions.
+
+- `Images-kws_asr.txt`: Tells the FPGA which memory regions to use for loading the binaries in sectors/** folder.
+
+### Add custom input
+
+The application performs inference on data found in the folder set by the CMake parameter `kws_asr_FILE_PATH`.
+
+To run the application with your own audio clips first create a folder to hold them and then copy the custom files into
+this folder:
+
+```commandline
+mkdir /tmp/custom_files
+
+cp custom_audio1.wav /tmp/custom_files/
+```
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+Next set `kws_asr_FILE_PATH` to the location of this folder when building:
+
+```commandline
+cmake \
+    -Dkws_asr_FILE_PATH=/tmp/custom_files/ \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=kws_asr- ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+The files found in the `kws_asr_FILE_PATH` folder will be picked up and automatically converted to C++ files during the
+CMake configuration stage and then compiled into the application during the build phase for performing inference with.
+
+The log from the configuration stage should tell you what directory path has been used:
+
+```log
+-- User option kws_asr_FILE_PATH is set to /tmp/custom_files
+```
+
+After compiling, your custom inputs will have now replaced the default ones in the application.
+
+### Add custom model
+
+The application performs KWS inference using the model pointed to by the CMake parameter `kws_asr_MODEL_TFLITE_PATH_KWS` and
+ASR inference using the model pointed to by the CMake parameter `kws_asr_MODEL_TFLITE_PATH_ASR`.
+
+This section assumes you wish to change the existing ASR model to a custom one. If instead you wish to change the KWS
+model then the instructions will be the same except ASR will change to KWS.
+
+> **Note:** If you want to run the model using Ethos-U55, ensure your custom model has been run through the Vela compiler successfully before continuing. See [Optimize model with Vela compiler](../sections/building.md#Optimize-custom-model-with-Vela-compiler).
+
+To run the application with a custom model you will need to provide a labels_<model_name>.txt file of labels
+associated with the model. Each line of the file should correspond to one of the outputs in your model. See the provided
+labels_wav2letter.txt file for an example.
+
+Then, you must set `kws_asr_MODEL_TFLITE_PATH_ASR` to the location of the Vela processed model file and
+`kws_asr_LABELS_TXT_FILE_ASR` to the location of the associated labels file.
+
+An example:
+
+```commandline
+cmake \
+    -Dkws_asr_MODEL_TFLITE_PATH_ASR=<path/to/custom_asr_model_after_vela.tflite> \
+    -Dkws_asr_LABELS_TXT_FILE_ASR=<path/to/labels_custom_model.txt> \
+    -DTARGET_PLATFORM=mps3 \
+    -DTARGET_SUBSYSTEM=sse-300 \
+    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/bare-metal-toolchain.cmake \
+    -DUSE_CASE_BUILD=kws_asr ..
+```
+
+For Windows, add `-G "MinGW Makefiles"` to the CMake command.
+
+> **Note:** Clean the build directory before re-running the CMake command.
+
+The `.tflite` model files pointed to by `kws_asr_MODEL_TFLITE_PATH_KWS` and `kws_asr_MODEL_TFLITE_PATH_ASR`, labels text files pointed to by `kws_asr_LABELS_TXT_FILE_KWS` and `kws_asr_LABELS_TXT_FILE_ASR`
+will be converted to C++ files during the CMake configuration stage and then compiled into the application for
+performing inference with.
+
+The log from the configuration stage should tell you what model path and labels file have been used:
+
+```log
+-- User option TARGET_PLATFORM is set to mps3
+-- User option kws_asr_MODEL_TFLITE_PATH_ASR is set to <path/to/custom_asr_model_after_vela.tflite>
+...
+-- User option kws_asr_LABELS_TXT_FILE_ASR is set to <path/to/labels_custom_model.txt>
+...
+-- Using <path/to/custom_asr_model_after_vela.tflite>
+++ Converting custom_asr_model_after_vela.tflite to\
+custom_asr_model_after_vela.tflite.cc
+-- Generating labels file from <path/to/labels_custom_model.txt>
+-- writing to Labels_wav2letter
+...
+```
+
+After compiling, your custom model will have now replaced the default one in the application.
+
+## Setting-up and running Ethos-U55 Code Samples
+
+### Setting up the Ethos-U55 Fast Model
+
+The FVP is available publicly from [Arm Ecosystem FVP downloads](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps).
+
+For Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains Ethos-U55 and
+Cortex-M55. The model is currently only supported on Linux based machines. To install the FVP:
+
+- Unpack the archive
+
+- Run the install script in the extracted package
+
+```commandline
+./FVP_Corstone_SSE-300_Ethos-U55.sh
+```
+
+- Follow the instructions to install the FVP to your desired location
+
+### Starting Fast Model simulation
+
+Once completed the building step, application binary ethos-u-kws_asr.axf can be found in the `build/bin` folder.
+Assuming the install location of the FVP was set to ~/FVP_install_location, the simulation can be started by:
+
+```commandline
+$ ~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55
+./bin/mps3-sse-300/ethos-u-kws_asr.axf
+```
+
+A log output should appear on the terminal:
+
+```log
+telnetterminal0: Listening for serial connection on port 5000
+telnetterminal1: Listening for serial connection on port 5001
+telnetterminal2: Listening for serial connection on port 5002
+telnetterminal5: Listening for serial connection on port 5003
+```
+
+This will also launch a telnet window with the sample application's standard output and error log entries containing
+information about the pre-built application version, TensorFlow Lite Micro library version used, data type as well as
+the input and output tensor sizes of the model compiled into the executable binary.
+
+After the application has started if `kws_asr_FILE_PATH` pointed to a single file (or a folder containing a single input file)
+the inference starts immediately. In case of multiple inputs choice, it outputs a menu and waits for the user input from telnet terminal:
+
+```log
+User input required
+Enter option number from:
+
+1. Classify next audio clip
+2. Classify audio clip at chosen index
+3. Run classification on all audio clips
+4. Show NN model info
+5. List audio clips
+
+Choice:
+
+```
+
+1. “Classify next audio clip” menu option will run single inference on the next included file.
+
+2. “Classify audio clip at chosen index” menu option will run inference on the chosen audio clip.
+
+    > **Note:** Please make sure to select audio clip index in the range of supplied audio clips during application build.
+
+3. “Run ... on all” menu option triggers sequential inference executions on all built-in files.
+
+4. “Show NN model info” menu option prints information about model data type, input and output tensor sizes:
+
+    ```log
+    [INFO] uTFL version: 2.5.0
+    [INFO] Model INPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 490 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:   1
+    [INFO] 		2:  49
+    [INFO] 		3:  10
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 1.107164
+    [INFO] ZeroPoint[0] = 95
+    [INFO] Model OUTPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 12 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:  12
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.003906
+    [INFO] ZeroPoint[0] = -128
+    [INFO] Activation buffer (a.k.a tensor arena) size used: 123616
+    [INFO] Number of operators: 16
+    [INFO] 	Operator 0: RESHAPE
+    [INFO] 	Operator 1: CONV_2D
+    [INFO] 	Operator 2: DEPTHWISE_CONV_2D
+    [INFO] 	Operator 3: CONV_2D
+    [INFO] 	Operator 4: DEPTHWISE_CONV_2D
+    [INFO] 	Operator 5: CONV_2D
+    [INFO] 	Operator 6: DEPTHWISE_CONV_2D
+    [INFO] 	Operator 7: CONV_2D
+    [INFO] 	Operator 8: DEPTHWISE_CONV_2D
+    [INFO] 	Operator 9: CONV_2D
+    [INFO] 	Operator 10: DEPTHWISE_CONV_2D
+    [INFO] 	Operator 11: CONV_2D
+    [INFO] 	Operator 12: AVERAGE_POOL_2D
+    [INFO] 	Operator 13: RESHAPE
+    [INFO] 	Operator 14: FULLY_CONNECTED
+    [INFO] 	Operator 15: SOFTMAX
+    [INFO] Model INPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 11544 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1: 296
+    [INFO] 		2:  39
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.110316
+    [INFO] ZeroPoint[0] = -11
+    [INFO] Model OUTPUT tensors:
+    [INFO] 	tensor type is INT8
+    [INFO] 	tensor occupies 4292 bytes with dimensions
+    [INFO] 		0:   1
+    [INFO] 		1:   1
+    [INFO] 		2: 148
+    [INFO] 		3:  29
+    [INFO] Quant dimension: 0
+    [INFO] Scale[0] = 0.003906
+    [INFO] ZeroPoint[0] = -128
+    [INFO] Activation buffer (a.k.a tensor arena) size used: 809808
+    [INFO] Number of operators: 1
+    [INFO] 	Operator 0: ethos-u
+    ```
+
+5. “List” menu option prints a list of pair ... indexes - the original filenames embedded in the application:
+
+    ```log
+    [INFO] List of Files:
+    [INFO] 0 => yesnogostop.wav
+    ```
+
+### Running Keyword Spotting and Automatic Speech Recognition
+
+Please select the first menu option to execute Keyword Spotting and Automatic Speech Recognition.
+
+The following example illustrates application output:
+
+```log
+[INFO] KWS audio data window size 16000
+[INFO] Running KWS inference on audio clip 0 => yesnogostop.wav
+[INFO] Inference 1/7
+[INFO] Profile for Inference:
+	Active NPU cycles: 0
+	Idle NPU cycles:   6
+
+[INFO] For timestamp: 0.000000 (inference #: 0); threshold: 0.900000
+[INFO] 		label @ 0: yes, score: 0.996094
+[INFO] Keyword spotted
+[INFO] Inference 1/2
+[INFO] Profile for Inference:
+	Active NPU cycles: 28924742
+	Idle NPU cycles:   424
+
+[INFO] Inference 2/2
+[INFO] Profile for Inference:
+	Active NPU cycles: 28924740
+	Idle NPU cycles:   426
+
+[INFO] Result for inf 0: no gow
+[INFO] Result for inf 1:  stoppe
+[INFO] Final result: no gow stoppe
+```
+
+It could take several minutes to complete one inference run (average time is 2-3 minutes).
+
+Using the input “yesnogostop.wav”, the log shows inference results for the KWS operation first, detecting the
+trigger word “yes“ with the stated probability score (in this case 0.99). After this, the ASR inference is run,
+printing the words recognized from the input sample.
+
+The profiling section of the log shows that for the ASR inference:
+
+- Ethos-U55's PMU report:
+
+  - 28,924,740 active cycles: number of cycles that were used for computation
+
+  - 426 idle cycles: number of cycles for which the NPU was idle
+
+- For FPGA platforms, CPU cycle count can also be enabled. For FVP, however, CPU cycle counters should not be used as
+the CPU model is not cycle-approximate or cycle-accurate.
+
+    Note that in this example the KWS inference does not use the Ethos-U55 and is run purely on CPU, therefore 0 Active
+    NPU cycles is shown.