summaryrefslogtreecommitdiff
path: root/source/use_case/ad
diff options
context:
space:
mode:
Diffstat (limited to 'source/use_case/ad')
-rw-r--r--source/use_case/ad/include/AdMelSpectrogram.hpp2
-rw-r--r--source/use_case/ad/include/AdModel.hpp2
-rw-r--r--source/use_case/ad/include/AdPostProcessing.hpp2
-rw-r--r--source/use_case/ad/include/MelSpectrogram.hpp61
-rw-r--r--source/use_case/ad/src/AdModel.cc14
-rw-r--r--source/use_case/ad/src/MelSpectrogram.cc114
-rw-r--r--source/use_case/ad/src/UseCaseHandler.cc13
7 files changed, 105 insertions, 103 deletions
diff --git a/source/use_case/ad/include/AdMelSpectrogram.hpp b/source/use_case/ad/include/AdMelSpectrogram.hpp
index 30a77c1..05c5bfc 100644
--- a/source/use_case/ad/include/AdMelSpectrogram.hpp
+++ b/source/use_case/ad/include/AdMelSpectrogram.hpp
@@ -69,7 +69,7 @@ namespace audio {
* energies to logarithmic scale. The difference from
* default behaviour is that the power is converted to dB
* and subsequently clamped.
- * @param[in/out] melEnergies - 1D vector of Mel energies
+ * @param[in,out] melEnergies - 1D vector of Mel energies
**/
virtual void ConvertToLogarithmicScale(std::vector<float>& melEnergies) override;
diff --git a/source/use_case/ad/include/AdModel.hpp b/source/use_case/ad/include/AdModel.hpp
index bbdf91c..8d914c4 100644
--- a/source/use_case/ad/include/AdModel.hpp
+++ b/source/use_case/ad/include/AdModel.hpp
@@ -44,7 +44,7 @@ namespace app {
static constexpr int ms_maxOpCnt = 6;
/* A mutable op resolver instance */
- tflite::MicroMutableOpResolver<ms_maxOpCnt> _m_opResolver;
+ tflite::MicroMutableOpResolver<ms_maxOpCnt> m_opResolver;
};
} /* namespace app */
diff --git a/source/use_case/ad/include/AdPostProcessing.hpp b/source/use_case/ad/include/AdPostProcessing.hpp
index f3b35a1..7eaec84 100644
--- a/source/use_case/ad/include/AdPostProcessing.hpp
+++ b/source/use_case/ad/include/AdPostProcessing.hpp
@@ -38,7 +38,7 @@ namespace app {
/** @brief Given a wav file name return AD model output index.
* @param[in] wavFileName Audio WAV filename.
- * File name should be in format <anything>_<goes>_XX_<here>.wav
+ * File name should be in format anything_goes_XX_here.wav
* where XX is the machine ID e.g. 00, 02, 04 or 06
* @return AD model output index as 8 bit integer.
**/
diff --git a/source/use_case/ad/include/MelSpectrogram.hpp b/source/use_case/ad/include/MelSpectrogram.hpp
index 22b5d29..d3ea3f7 100644
--- a/source/use_case/ad/include/MelSpectrogram.hpp
+++ b/source/use_case/ad/include/MelSpectrogram.hpp
@@ -65,16 +65,16 @@ namespace audio {
/**
* @brief Extract Mel Spectrogram for one single small frame of
* audio data e.g. 640 samples.
- * @param[in] audioData - Vector of audio samples to calculate
+ * @param[in] audioData Vector of audio samples to calculate
* features for.
- * @param[in] trainingMean - Value to subtract from the the computed mel spectrogram, default 0.
+ * @param[in] trainingMean Value to subtract from the the computed mel spectrogram, default 0.
* @return Vector of extracted Mel Spectrogram features.
**/
std::vector<float> ComputeMelSpec(const std::vector<int16_t>& audioData, float trainingMean = 0);
/**
* @brief Constructor
- * @param[in] params - Mel Spectrogram parameters
+ * @param[in] params Mel Spectrogram parameters
*/
explicit MelSpectrogram(const MelSpecParams& params);
@@ -87,10 +87,11 @@ namespace audio {
/**
* @brief Extract Mel Spectrogram features and quantise for one single small
* frame of audio data e.g. 640 samples.
- * @param[in] audioData - Vector of audio samples to calculate
+ * @param[in] audioData Vector of audio samples to calculate
* features for.
- * @param[in] quantScale - quantisation scale.
- * @param[in] quantOffset - quantisation offset
+ * @param[in] quantScale quantisation scale.
+ * @param[in] quantOffset quantisation offset.
+ * @param[in] trainingMean training mean.
* @return Vector of extracted quantised Mel Spectrogram features.
**/
template<typename T>
@@ -103,12 +104,12 @@ namespace audio {
float minVal = std::numeric_limits<T>::min();
float maxVal = std::numeric_limits<T>::max();
- std::vector<T> melSpecOut(this->_m_params.m_numFbankBins);
- const size_t numFbankBins = this->_m_params.m_numFbankBins;
+ std::vector<T> melSpecOut(this->m_params.m_numFbankBins);
+ const size_t numFbankBins = this->m_params.m_numFbankBins;
/* Quantize to T. */
for (size_t k = 0; k < numFbankBins; ++k) {
- auto quantizedEnergy = std::round(((this->_m_melEnergies[k]) / quantScale) + quantOffset);
+ auto quantizedEnergy = std::round(((this->m_melEnergies[k]) / quantScale) + quantOffset);
melSpecOut[k] = static_cast<T>(std::min<float>(std::max<float>(quantizedEnergy, minVal), maxVal));
}
@@ -124,9 +125,9 @@ namespace audio {
protected:
/**
* @brief Project input frequency to Mel Scale.
- * @param[in] freq - input frequency in floating point
- * @param[in] useHTKmethod - bool to signal if HTK method is to be
- * used for calculation
+ * @param[in] freq input frequency in floating point
+ * @param[in] useHTKMethod bool to signal if HTK method is to be
+ * used for calculation
* @return Mel transformed frequency in floating point
**/
static float MelScale(const float freq,
@@ -135,9 +136,9 @@ namespace audio {
/**
* @brief Inverse Mel transform - convert MEL warped frequency
* back to normal frequency
- * @param[in] freq - Mel frequency in floating point
- * @param[in] useHTKmethod - bool to signal if HTK method is to be
- * used for calculation
+ * @param[in] melFreq Mel frequency in floating point
+ * @param[in] useHTKMethod bool to signal if HTK method is to be
+ * used for calculation
* @return Real world frequency in floating point
**/
static float InverseMelScale(const float melFreq,
@@ -168,7 +169,7 @@ namespace audio {
/**
* @brief Converts the Mel energies for logarithmic scale
- * @param[in/out] melEnergies - 1D vector of Mel energies
+ * @param[in,out] melEnergies 1D vector of Mel energies
**/
virtual void ConvertToLogarithmicScale(std::vector<float>& melEnergies);
@@ -176,10 +177,10 @@ namespace audio {
* @brief Given the low and high Mel values, get the normaliser
* for weights to be applied when populating the filter
* bank.
- * @param[in] leftMel - low Mel frequency value
- * @param[in] rightMel - high Mel frequency value
- * @param[in] useHTKMethod - bool to signal if HTK method is to be
- * used for calculation
+ * @param[in] leftMel low Mel frequency value
+ * @param[in] rightMel high Mel frequency value
+ * @param[in] useHTKMethod bool to signal if HTK method is to be
+ * used for calculation
* @return Return float value to be applied
* when populating the filter bank.
*/
@@ -189,16 +190,16 @@ namespace audio {
const bool useHTKMethod);
private:
- MelSpecParams _m_params;
- std::vector<float> _m_frame;
- std::vector<float> _m_buffer;
- std::vector<float> _m_melEnergies;
- std::vector<float> _m_windowFunc;
- std::vector<std::vector<float>> _m_melFilterBank;
- std::vector<uint32_t> _m_filterBankFilterFirst;
- std::vector<uint32_t> _m_filterBankFilterLast;
- bool _m_filterBankInitialised;
- arm::app::math::FftInstance _m_fftInstance;
+ MelSpecParams m_params;
+ std::vector<float> m_frame;
+ std::vector<float> m_buffer;
+ std::vector<float> m_melEnergies;
+ std::vector<float> m_windowFunc;
+ std::vector<std::vector<float>> m_melFilterBank;
+ std::vector<uint32_t> m_filterBankFilterFirst;
+ std::vector<uint32_t> m_filterBankFilterLast;
+ bool m_filterBankInitialised;
+ arm::app::math::FftInstance m_fftInstance;
/**
* @brief Initialises the filter banks.
diff --git a/source/use_case/ad/src/AdModel.cc b/source/use_case/ad/src/AdModel.cc
index 148bc98..82ad822 100644
--- a/source/use_case/ad/src/AdModel.cc
+++ b/source/use_case/ad/src/AdModel.cc
@@ -20,19 +20,19 @@
const tflite::MicroOpResolver& arm::app::AdModel::GetOpResolver()
{
- return this->_m_opResolver;
+ return this->m_opResolver;
}
bool arm::app::AdModel::EnlistOperations()
{
- this->_m_opResolver.AddAveragePool2D();
- this->_m_opResolver.AddConv2D();
- this->_m_opResolver.AddDepthwiseConv2D();
- this->_m_opResolver.AddRelu6();
- this->_m_opResolver.AddReshape();
+ this->m_opResolver.AddAveragePool2D();
+ this->m_opResolver.AddConv2D();
+ this->m_opResolver.AddDepthwiseConv2D();
+ this->m_opResolver.AddRelu6();
+ this->m_opResolver.AddReshape();
#if defined(ARM_NPU)
- if (kTfLiteOk == this->_m_opResolver.AddEthosU()) {
+ if (kTfLiteOk == this->m_opResolver.AddEthosU()) {
info("Added %s support to op resolver\n",
tflite::GetString_ETHOSU());
} else {
diff --git a/source/use_case/ad/src/MelSpectrogram.cc b/source/use_case/ad/src/MelSpectrogram.cc
index f1752e1..fa7714a 100644
--- a/source/use_case/ad/src/MelSpectrogram.cc
+++ b/source/use_case/ad/src/MelSpectrogram.cc
@@ -61,27 +61,27 @@ namespace audio {
}
MelSpectrogram::MelSpectrogram(const MelSpecParams& params):
- _m_params(params),
- _m_filterBankInitialised(false)
+ m_params(params),
+ m_filterBankInitialised(false)
{
- this->_m_buffer = std::vector<float>(
- this->_m_params.m_frameLenPadded, 0.0);
- this->_m_frame = std::vector<float>(
- this->_m_params.m_frameLenPadded, 0.0);
- this->_m_melEnergies = std::vector<float>(
- this->_m_params.m_numFbankBins, 0.0);
+ this->m_buffer = std::vector<float>(
+ this->m_params.m_frameLenPadded, 0.0);
+ this->m_frame = std::vector<float>(
+ this->m_params.m_frameLenPadded, 0.0);
+ this->m_melEnergies = std::vector<float>(
+ this->m_params.m_numFbankBins, 0.0);
- this->_m_windowFunc = std::vector<float>(this->_m_params.m_frameLen);
- const auto multiplier = static_cast<float>(2 * M_PI / this->_m_params.m_frameLen);
+ this->m_windowFunc = std::vector<float>(this->m_params.m_frameLen);
+ const auto multiplier = static_cast<float>(2 * M_PI / this->m_params.m_frameLen);
/* Create window function. */
- for (size_t i = 0; i < this->_m_params.m_frameLen; ++i) {
- this->_m_windowFunc[i] = (0.5 - (0.5 *
+ for (size_t i = 0; i < this->m_params.m_frameLen; ++i) {
+ this->m_windowFunc[i] = (0.5 - (0.5 *
math::MathUtils::CosineF32(static_cast<float>(i) * multiplier)));
}
- math::MathUtils::FftInitF32(this->_m_params.m_frameLenPadded, this->_m_fftInstance);
- debug("Instantiated Mel Spectrogram object: %s\n", this->_m_params.Str().c_str());
+ math::MathUtils::FftInitF32(this->m_params.m_frameLenPadded, this->m_fftInstance);
+ debug("Instantiated Mel Spectrogram object: %s\n", this->m_params.Str().c_str());
}
void MelSpectrogram::Init()
@@ -161,20 +161,20 @@ namespace audio {
void MelSpectrogram::ConvertToPowerSpectrum()
{
- const uint32_t halfDim = this->_m_buffer.size() / 2;
+ const uint32_t halfDim = this->m_buffer.size() / 2;
/* Handle this special case. */
- float firstEnergy = this->_m_buffer[0] * this->_m_buffer[0];
- float lastEnergy = this->_m_buffer[1] * this->_m_buffer[1];
+ float firstEnergy = this->m_buffer[0] * this->m_buffer[0];
+ float lastEnergy = this->m_buffer[1] * this->m_buffer[1];
math::MathUtils::ComplexMagnitudeSquaredF32(
- this->_m_buffer.data(),
- this->_m_buffer.size(),
- this->_m_buffer.data(),
- this->_m_buffer.size()/2);
+ this->m_buffer.data(),
+ this->m_buffer.size(),
+ this->m_buffer.data(),
+ this->m_buffer.size()/2);
- this->_m_buffer[0] = firstEnergy;
- this->_m_buffer[halfDim] = lastEnergy;
+ this->m_buffer[0] = firstEnergy;
+ this->m_buffer[halfDim] = lastEnergy;
}
float MelSpectrogram::GetMelFilterBankNormaliser(
@@ -193,14 +193,14 @@ namespace audio {
void MelSpectrogram::InitMelFilterBank()
{
if (!this->IsMelFilterBankInited()) {
- this->_m_melFilterBank = this->CreateMelFilterBank();
- this->_m_filterBankInitialised = true;
+ this->m_melFilterBank = this->CreateMelFilterBank();
+ this->m_filterBankInitialised = true;
}
}
bool MelSpectrogram::IsMelFilterBankInited() const
{
- return this->_m_filterBankInitialised;
+ return this->m_filterBankInitialised;
}
std::vector<float> MelSpectrogram::ComputeMelSpec(const std::vector<int16_t>& audioData, float trainingMean)
@@ -209,64 +209,64 @@ namespace audio {
/* TensorFlow way of normalizing .wav data to (-1, 1). */
constexpr float normaliser = 1.0/(1<<15);
- for (size_t i = 0; i < this->_m_params.m_frameLen; ++i) {
- this->_m_frame[i] = static_cast<float>(audioData[i]) * normaliser;
+ for (size_t i = 0; i < this->m_params.m_frameLen; ++i) {
+ this->m_frame[i] = static_cast<float>(audioData[i]) * normaliser;
}
/* Apply window function to input frame. */
- for(size_t i = 0; i < this->_m_params.m_frameLen; ++i) {
- this->_m_frame[i] *= this->_m_windowFunc[i];
+ for(size_t i = 0; i < this->m_params.m_frameLen; ++i) {
+ this->m_frame[i] *= this->m_windowFunc[i];
}
/* Set remaining frame values to 0. */
- std::fill(this->_m_frame.begin() + this->_m_params.m_frameLen,this->_m_frame.end(), 0);
+ std::fill(this->m_frame.begin() + this->m_params.m_frameLen,this->m_frame.end(), 0);
/* Compute FFT. */
- math::MathUtils::FftF32(this->_m_frame, this->_m_buffer, this->_m_fftInstance);
+ math::MathUtils::FftF32(this->m_frame, this->m_buffer, this->m_fftInstance);
/* Convert to power spectrum. */
this->ConvertToPowerSpectrum();
/* Apply mel filterbanks. */
- if (!this->ApplyMelFilterBank(this->_m_buffer,
- this->_m_melFilterBank,
- this->_m_filterBankFilterFirst,
- this->_m_filterBankFilterLast,
- this->_m_melEnergies)) {
+ if (!this->ApplyMelFilterBank(this->m_buffer,
+ this->m_melFilterBank,
+ this->m_filterBankFilterFirst,
+ this->m_filterBankFilterLast,
+ this->m_melEnergies)) {
printf_err("Failed to apply MEL filter banks\n");
}
/* Convert to logarithmic scale */
- this->ConvertToLogarithmicScale(this->_m_melEnergies);
+ this->ConvertToLogarithmicScale(this->m_melEnergies);
/* Perform mean subtraction. */
- for (auto& energy:this->_m_melEnergies) {
+ for (auto& energy:this->m_melEnergies) {
energy -= trainingMean;
}
- return this->_m_melEnergies;
+ return this->m_melEnergies;
}
std::vector<std::vector<float>> MelSpectrogram::CreateMelFilterBank()
{
- size_t numFftBins = this->_m_params.m_frameLenPadded / 2;
- float fftBinWidth = static_cast<float>(this->_m_params.m_samplingFreq) / this->_m_params.m_frameLenPadded;
+ size_t numFftBins = this->m_params.m_frameLenPadded / 2;
+ float fftBinWidth = static_cast<float>(this->m_params.m_samplingFreq) / this->m_params.m_frameLenPadded;
- float melLowFreq = MelSpectrogram::MelScale(this->_m_params.m_melLoFreq,
- this->_m_params.m_useHtkMethod);
- float melHighFreq = MelSpectrogram::MelScale(this->_m_params.m_melHiFreq,
- this->_m_params.m_useHtkMethod);
- float melFreqDelta = (melHighFreq - melLowFreq) / (this->_m_params.m_numFbankBins + 1);
+ float melLowFreq = MelSpectrogram::MelScale(this->m_params.m_melLoFreq,
+ this->m_params.m_useHtkMethod);
+ float melHighFreq = MelSpectrogram::MelScale(this->m_params.m_melHiFreq,
+ this->m_params.m_useHtkMethod);
+ float melFreqDelta = (melHighFreq - melLowFreq) / (this->m_params.m_numFbankBins + 1);
std::vector<float> thisBin = std::vector<float>(numFftBins);
std::vector<std::vector<float>> melFilterBank(
- this->_m_params.m_numFbankBins);
- this->_m_filterBankFilterFirst =
- std::vector<uint32_t>(this->_m_params.m_numFbankBins);
- this->_m_filterBankFilterLast =
- std::vector<uint32_t>(this->_m_params.m_numFbankBins);
+ this->m_params.m_numFbankBins);
+ this->m_filterBankFilterFirst =
+ std::vector<uint32_t>(this->m_params.m_numFbankBins);
+ this->m_filterBankFilterLast =
+ std::vector<uint32_t>(this->m_params.m_numFbankBins);
- for (size_t bin = 0; bin < this->_m_params.m_numFbankBins; bin++) {
+ for (size_t bin = 0; bin < this->m_params.m_numFbankBins; bin++) {
float leftMel = melLowFreq + bin * melFreqDelta;
float centerMel = melLowFreq + (bin + 1) * melFreqDelta;
float rightMel = melLowFreq + (bin + 2) * melFreqDelta;
@@ -274,11 +274,11 @@ namespace audio {
uint32_t firstIndex = 0;
uint32_t lastIndex = 0;
bool firstIndexFound = false;
- const float normaliser = this->GetMelFilterBankNormaliser(leftMel, rightMel, this->_m_params.m_useHtkMethod);
+ const float normaliser = this->GetMelFilterBankNormaliser(leftMel, rightMel, this->m_params.m_useHtkMethod);
for (size_t i = 0; i < numFftBins; ++i) {
float freq = (fftBinWidth * i); /* Center freq of this fft bin. */
- float mel = MelSpectrogram::MelScale(freq, this->_m_params.m_useHtkMethod);
+ float mel = MelSpectrogram::MelScale(freq, this->m_params.m_useHtkMethod);
thisBin[i] = 0.0;
if (mel > leftMel && mel < rightMel) {
@@ -298,8 +298,8 @@ namespace audio {
}
}
- this->_m_filterBankFilterFirst[bin] = firstIndex;
- this->_m_filterBankFilterLast[bin] = lastIndex;
+ this->m_filterBankFilterFirst[bin] = firstIndex;
+ this->m_filterBankFilterLast[bin] = lastIndex;
/* Copy the part we care about. */
for (uint32_t i = firstIndex; i <= lastIndex; ++i) {
diff --git a/source/use_case/ad/src/UseCaseHandler.cc b/source/use_case/ad/src/UseCaseHandler.cc
index 233b0f4..ec35156 100644
--- a/source/use_case/ad/src/UseCaseHandler.cc
+++ b/source/use_case/ad/src/UseCaseHandler.cc
@@ -30,13 +30,13 @@ namespace app {
/**
* @brief Helper function to increment current audio clip index
- * @param[in/out] ctx pointer to the application context object
+ * @param[in,out] ctx pointer to the application context object
**/
static void IncrementAppCtxClipIdx(ApplicationContext& ctx);
/**
* @brief Helper function to set the audio clip index
- * @param[in/out] ctx pointer to the application context object
+ * @param[in,out] ctx pointer to the application context object
* @param[in] idx value to be set
* @return true if index is set, false otherwise
**/
@@ -47,7 +47,7 @@ namespace app {
* object.
* @param[in] platform reference to the hal platform object
* @param[in] result average sum of classification results
- * @param[in] threhsold if larger than this value we have an anomaly
+ * @param[in] threshold if larger than this value we have an anomaly
* @return true if successful, false otherwise
**/
static bool PresentInferenceResult(hal_platform& platform, float result, float threshold);
@@ -61,9 +61,10 @@ namespace app {
*
* Warning: mfcc calculator provided as input must have the same life scope as returned function.
*
- * @param[in] mfcc MFCC feature calculator.
- * @param[in/out] inputTensor Input tensor pointer to store calculated features.
- * @param[i] cacheSize Size of the feture vectors cache (number of feature vectors).
+ * @param[in] melSpec MFCC feature calculator.
+ * @param[in,out] inputTensor Input tensor pointer to store calculated features.
+ * @param[in] cacheSize Size of the feture vectors cache (number of feature vectors).
+ * @param[in] trainingMean Training mean.
* @return function function to be called providing audio sample and sliding window index.
*/
static std::function<void (std::vector<int16_t>&, int, bool, size_t, size_t)>
generated by cgit v1.2.1 (git 2.23.0) at 2024-06-26 01:51:19 +0000