MLECO-1252 ASR sample application using the public ArmNN C++ API.

Change-Id: I98cd505b8772a8c8fa88308121bc94135bb45068
Signed-off-by: Éanna Ó Catháin <eanna.ocathain@arm.com>

4 files changed, 385 insertions, 0 deletions

diff --git a/samples/SpeechRecognition/test/AudioCaptureTest.cpp b/samples/SpeechRecognition/test/AudioCaptureTest.cpp
new file mode 100644
index 0000000000..94b4e7cb7a
--- /dev/null
+++ b/samples/SpeechRecognition/test/AudioCaptureTest.cpp
@@ -0,0 +1,61 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#define CATCH_CONFIG_MAIN
+#include <catch.hpp>
+#include <limits>
+
+#include "AudioCapture.hpp"
+
+TEST_CASE("Test capture of audio file")
+{
+    std::string testResources = TEST_RESOURCE_DIR;
+    REQUIRE(testResources != "");
+    std::string file =  testResources + "/" + "myVoiceIsMyPassportVerifyMe04.wav";
+    asr::AudioCapture capture;
+    std::vector<float> audioData = capture.LoadAudioFile(file);
+    capture.InitSlidingWindow(audioData.data(), audioData.size(), 47712, 16000);
+
+    std::vector<float> firstAudioBlock = capture.Next();
+    float actual1 = firstAudioBlock.at(0);
+    float actual2 = firstAudioBlock.at(47000);
+    CHECK(std::to_string(actual1) == "0.000352");
+    CHECK(std::to_string(actual2) == "-0.056441");
+    CHECK(firstAudioBlock.size() == 47712);
+
+    CHECK(capture.HasNext() == true);
+
+    std::vector<float> secondAudioBlock = capture.Next();
+    float actual3 = secondAudioBlock.at(0);
+    float actual4 = secondAudioBlock.at(47000);
+    CHECK(std::to_string(actual3) == "0.102077");
+    CHECK(std::to_string(actual4) == "0.000194");
+    CHECK(capture.HasNext() == true);
+
+    std::vector<float> thirdAudioBlock = capture.Next();
+    float actual5 = thirdAudioBlock.at(0);
+    float actual6 = thirdAudioBlock.at(33500);
+    float actual7 = thirdAudioBlock.at(33600);
+    CHECK(std::to_string(actual5) == "-0.076416");
+    CHECK(std::to_string(actual6) == "-0.000275");
+    CHECK(std::to_string(actual7) == "0.000000");
+    CHECK(capture.HasNext() == false);
+}
+
+TEST_CASE("Test sliding window of audio capture")
+{
+    std::string testResources = TEST_RESOURCE_DIR;
+    REQUIRE(testResources != "");
+    std::string file =  testResources + "/" + "myVoiceIsMyPassportVerifyMe04.wav";
+    asr::AudioCapture capture;
+    std::vector<float> audioData = capture.LoadAudioFile(file);
+    capture.InitSlidingWindow(audioData.data(), audioData.size(), 47712, 16000);
+    capture.Next();
+    capture.Next();
+
+    CHECK(capture.HasNext() == true);
+    capture.Next();
+    CHECK(capture.HasNext() == false);
+}
diff --git a/samples/SpeechRecognition/test/DecoderTest.cpp b/samples/SpeechRecognition/test/DecoderTest.cpp
new file mode 100644
index 0000000000..13a3905b99
--- /dev/null
+++ b/samples/SpeechRecognition/test/DecoderTest.cpp
@@ -0,0 +1,86 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include <catch.hpp>
+#include <map>
+#include "Decoder.hpp"
+
+std::map<int, std::string> labels = {
+        {0, "a" },
+        {1, "b" },
+        {2, "c" },
+        {3, "d" },
+        {4, "e" },
+        {5, "f" },
+        {6, "g" },
+        {7, "h" },
+        {8, "i" },
+        {9, "j" },
+        {10,"k" },
+        {11,"l" },
+        {12,"m" },
+        {13,"n" },
+        {14,"o" },
+        {15,"p" },
+        {16,"q" },
+        {17,"r" },
+        {18,"s" },
+        {19,"t" },
+        {20,"u" },
+        {21,"v" },
+        {22,"w" },
+        {23,"x" },
+        {24,"y" },
+        {25,"z" },
+        {26, "\'" },
+        {27, " "},
+        {28,"$" }
+};
+
+TEST_CASE("Test Wav2Letter output decoder")
+{
+
+    std::vector<uint16_t> outputValues =
+            {
+            1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2,
+
+            1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2
+            };
+
+    std::vector<int8_t> convertedValues;
+
+    for(uint16_t outputVal : outputValues)
+    {
+        convertedValues.emplace_back(static_cast<int8_t>(outputVal));
+    }
+
+    asr::Decoder decoder(labels);
+    std::string text = decoder.DecodeOutput<int8_t>(convertedValues);
+    CHECK(text == "hello");
+}
+
+
diff --git a/samples/SpeechRecognition/test/MFCCTest.cpp b/samples/SpeechRecognition/test/MFCCTest.cpp
new file mode 100644
index 0000000000..2a552643d5
--- /dev/null
+++ b/samples/SpeechRecognition/test/MFCCTest.cpp
@@ -0,0 +1,102 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include <catch.hpp>
+#include <limits>
+
+#include "MFCC.hpp"
+
+const std::vector<float> testWav = std::vector<float>{
+    -3.0f, 0.0f, 1.0f, -1.0f, 2.0f, 3.0f, -2.0f, 2.0f,
+            1.0f, -2.0f, 0.0f, 3.0f, -1.0f, 8.0f, 3.0f, 2.0f,
+            -1.0f, -1.0f, 2.0f, 7.0f, 3.0f, 5.0f, 6.0f, 6.0f,
+            6.0f, 12.0f, 5.0f, 6.0f, 3.0f, 3.0f, 5.0f, 4.0f,
+            4.0f, 6.0f, 7.0f, 7.0f, 7.0f, 3.0f, 7.0f, 2.0f,
+            8.0f, 4.0f, 4.0f, 2.0f, -4.0f, -1.0f, -1.0f, -4.0f,
+            2.0f, 1.0f, -1.0f, -4.0f, 0.0f, -7.0f, -6.0f, -2.0f,
+            -5.0f, 1.0f, -5.0f, -1.0f, -7.0f, -3.0f, -3.0f, -7.0f,
+            0.0f, -3.0f, 3.0f, -5.0f, 0.0f, 1.0f, -2.0f, -2.0f,
+            -3.0f, -3.0f, -7.0f, -3.0f, -2.0f, -6.0f, -5.0f, -8.0f,
+            -2.0f, -8.0f, 4.0f, -9.0f, -4.0f, -9.0f, -5.0f, -5.0f,
+            -3.0f, -9.0f, -3.0f, -9.0f, -1.0f, -7.0f, -4.0f, 1.0f,
+            -3.0f, 2.0f, -8.0f, -4.0f, -4.0f, -5.0f, 1.0f, -3.0f,
+            -1.0f, 0.0f, -1.0f, -2.0f, -3.0f, -2.0f, -4.0f, -1.0f,
+            1.0f, -1.0f, 3.0f, 0.0f, 3.0f, 2.0f, 0.0f, 0.0f,
+            0.0f, -3.0f, 1.0f, 1.0f, 0.0f, 8.0f, 3.0f, 4.0f,
+            1.0f, 5.0f, 6.0f, 4.0f, 7.0f, 3.0f, 3.0f, 0.0f,
+            3.0f, 6.0f, 7.0f, 6.0f, 4.0f, 5.0f, 9.0f, 9.0f,
+            5.0f, 5.0f, 8.0f, 1.0f, 6.0f, 9.0f, 6.0f, 6.0f,
+            7.0f, 1.0f, 8.0f, 1.0f, 5.0f, 0.0f, 5.0f, 5.0f,
+            0.0f, 3.0f, 2.0f, 7.0f, 2.0f, -3.0f, 3.0f, 0.0f,
+            3.0f, 0.0f, 0.0f, 0.0f, 2.0f, 0.0f, -1.0f, -1.0f,
+            -2.0f, -3.0f, -8.0f, 0.0f, 1.0f, 0.0f, -3.0f, -3.0f,
+            -3.0f, -2.0f, -3.0f, -3.0f, -4.0f, -6.0f, -2.0f, -8.0f,
+            -9.0f, -4.0f, -1.0f, -5.0f, -3.0f, -3.0f, -4.0f, -3.0f,
+            -6.0f, 3.0f, 0.0f, -1.0f, -2.0f, -9.0f, -4.0f, -2.0f,
+            2.0f, -1.0f, 3.0f, -5.0f, -5.0f, -2.0f, 0.0f, -2.0f,
+            0.0f, -1.0f, -3.0f, 1.0f, -2.0f, 9.0f, 4.0f, 5.0f,
+            2.0f, 2.0f, 1.0f, 0.0f, -6.0f, -2.0f, 0.0f, 0.0f,
+            0.0f, -1.0f, 4.0f, -4.0f, 3.0f, -7.0f, -1.0f, 5.0f,
+            -6.0f, -1.0f, -5.0f, 4.0f, 3.0f, 9.0f, -2.0f, 1.0f,
+            3.0f, 0.0f, 0.0f, -2.0f, 1.0f, 2.0f, 1.0f, 1.0f,
+            0.0f, 3.0f, 2.0f, -1.0f, 3.0f, -3.0f, 7.0f, 0.0f,
+            0.0f, 3.0f, 2.0f, 2.0f, -2.0f, 3.0f, -2.0f, 2.0f,
+            -3.0f, 4.0f, -1.0f, -1.0f, -5.0f, -1.0f, -3.0f, -2.0f,
+            1.0f, -1.0f, 3.0f, 2.0f, 4.0f, 1.0f, 2.0f, -2.0f,
+            0.0f, 2.0f, 7.0f, 0.0f, 8.0f, -3.0f, 6.0f, -3.0f,
+            6.0f, 1.0f, 2.0f, -3.0f, -1.0f, -1.0f, -1.0f, 1.0f,
+            -2.0f, 2.0f, 1.0f, 2.0f, 0.0f, -2.0f, 3.0f, -2.0f,
+            3.0f, -2.0f, 1.0f, 0.0f, -3.0f, -1.0f, -2.0f, -4.0f,
+            -6.0f, -5.0f, -8.0f, -1.0f, -4.0f, 0.0f, -3.0f, -1.0f,
+            -1.0f, -1.0f, 0.0f, -2.0f, -3.0f, -7.0f, -1.0f, 0.0f,
+            1.0f, 5.0f, 0.0f, 5.0f, 1.0f, 1.0f, -3.0f, 0.0f,
+            -6.0f, 3.0f, -8.0f, 4.0f, -8.0f, 6.0f, -6.0f, 1.0f,
+            -6.0f, -2.0f, -5.0f, -6.0f, 0.0f, -5.0f, 4.0f, -1.0f,
+            4.0f, -2.0f, 1.0f, 2.0f, 1.0f, 0.0f, -2.0f, 0.0f,
+            0.0f, 2.0f, -2.0f, 2.0f, -5.0f, 2.0f, 0.0f, -2.0f,
+            1.0f, -2.0f, 0.0f, 5.0f, 1.0f, 0.0f, 1.0f, 5.0f,
+            0.0f, 8.0f, 3.0f, 2.0f, 2.0f, 0.0f, 5.0f, -2.0f,
+            3.0f, 1.0f, 0.0f, 1.0f, 0.0f, -2.0f, -1.0f, -3.0f,
+            1.0f, -1.0f, 3.0f, 0.0f, 3.0f, 0.0f, -2.0f, -1.0f,
+            -4.0f, -4.0f, -4.0f, -1.0f, -4.0f, -4.0f, -3.0f, -6.0f,
+            -3.0f, -7.0f, -3.0f, -1.0f, -2.0f, 0.0f, -5.0f, -4.0f,
+            -7.0f, -3.0f, -2.0f, -2.0f, 1.0f, 2.0f, 2.0f, 8.0f,
+            5.0f, 4.0f, 2.0f, 4.0f, 3.0f, 5.0f, 0.0f, 3.0f,
+            3.0f, 6.0f, 4.0f, 2.0f, 2.0f, -2.0f, 4.0f, -2.0f,
+            3.0f, 3.0f, 2.0f, 1.0f, 1.0f, 4.0f, -5.0f, 2.0f,
+            -3.0f, 0.0f, -1.0f, 1.0f, -2.0f, 2.0f, 5.0f, 1.0f,
+            4.0f, 2.0f, 3.0f, 1.0f, -1.0f, 1.0f, 0.0f, 6.0f,
+            0.0f, -2.0f, -1.0f, 1.0f, -1.0f, 2.0f, -5.0f, -1.0f,
+            -5.0f, -1.0f, -6.0f, -3.0f, -3.0f, 2.0f, 4.0f, 0.0f,
+            -1.0f, -5.0f, 3.0f, -4.0f, -1.0f, -3.0f, -4.0f, 1.0f,
+            -4.0f, 1.0f, -1.0f, -1.0f, 0.0f, -5.0f, -4.0f, -2.0f,
+            -1.0f, -1.0f, -3.0f, -7.0f, -3.0f, -3.0f, 4.0f, 4.0f
+};
+
+TEST_CASE("Test MFCC")
+{
+    int sampFreq = 16000;
+    int frameLenMs = 32;
+    int frameLenSamples = sampFreq * frameLenMs * 0.001;
+    int numMfccFeats = 13;
+
+    std::vector<float> fullAudioData;
+
+        for (auto f : testWav)
+            {
+                fullAudioData.emplace_back( f / (1<<15));
+            }
+
+
+    MfccParams mfccParams(sampFreq, 128, 0, 8000, numMfccFeats, frameLenSamples, false, 1);
+
+    MFCC mfccInst = MFCC(mfccParams);
+    auto mfccOutput = mfccInst.MfccCompute(fullAudioData);
+
+    std::vector<float> expected = { -834.96564f, 21.02699f, 18.62856f, 7.3412f, 18.90791f, -5.36034f, 6.52351f,
+                                    -11.27064f, 8.37522f, 12.0672f, 8.30833f, -13.50008f, -18.1761f};
+
+    REQUIRE_THAT(mfccOutput, Catch::Approx(expected).epsilon(1.e-5) );
+}
\ No newline at end of file
diff --git a/samples/SpeechRecognition/test/PreprocessTest.cpp b/samples/SpeechRecognition/test/PreprocessTest.cpp
new file mode 100644
index 0000000000..2b98831fda
--- /dev/null
+++ b/samples/SpeechRecognition/test/PreprocessTest.cpp
@@ -0,0 +1,136 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include <catch.hpp>
+#include <limits>
+
+#include "Preprocess.hpp"
+#include "DataStructures.hpp"
+
+void PopulateTestWavVector(std::vector<int16_t>& vec)
+{
+    constexpr int int16max = std::numeric_limits<int16_t>::max();
+    int val = 0;
+    for (size_t i = 0; i < vec.size(); ++i, ++val)
+    {
+
+        /* We want a differential filter response from both - order 1
+         * and 2 => Don't have a linear signal here - we use a signal
+         * using squares for example. Alternate sign flips might work
+         * just as well and will be computationally less work! */
+        int valsq = val * val;
+        if (valsq > int16max)
+        {
+            val = 0;
+            valsq = 0;
+        }
+        vec[i] = valsq;
+    }
+}
+
+TEST_CASE("Preprocessing calculation INT8")
+{
+    /*Test  Constants: */
+    const uint32_t  windowLen             = 512;
+    const uint32_t  windowStride          = 160;
+    const float     quantScale            = 0.1410219967365265;
+    const int       quantOffset           = -11;
+    int             numMfccVectors        = 10;
+    const int       sampFreq              = 16000;
+    const int       frameLenMs            = 32;
+    const int       frameLenSamples       = sampFreq * frameLenMs * 0.001;
+    const int       numMfccFeats          = 13;
+    const int       audioDataToPreProcess = 512 + ((numMfccVectors -1) * windowStride);
+    int             outputBufferSize = numMfccVectors * numMfccFeats * 3;
+
+    /* Test wav memory */
+    std::vector <int16_t> testWav1((windowStride * numMfccVectors) +
+                              (windowLen - windowStride));
+    /* Populate with dummy input */
+    PopulateTestWavVector(testWav1);
+
+    MfccParams mfccParams(sampFreq, 128, 0, 8000, numMfccFeats, frameLenSamples, false, numMfccVectors);
+
+    MFCC mfccInst = MFCC(mfccParams);
+
+    std::vector<float> fullAudioData;
+
+    for(int i = 0; i < 4; ++i)
+    {
+        for (auto f : testWav1)
+        {
+            fullAudioData.emplace_back(static_cast<float>(f) / (1<<15));
+        }
+    }
+
+    Preprocess prep(frameLenSamples, windowStride, mfccInst);
+
+    std::vector<int8_t> outputBuffer(outputBufferSize);
+
+    prep.Invoke(fullAudioData.data(), audioDataToPreProcess, outputBuffer, quantOffset, quantScale);
+
+    int8_t expectedResult[numMfccVectors][numMfccFeats*3] =
+    {
+            /* Feature vec 0 */
+            -32, 4, -9, -8, -10, -10, -11, -11, -11, -11, -12, -11, -11,    /* MFCCs   */
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,    /* Delta 1 */
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,    /* Delta 2 */
+
+            /* Feature vec 1 */
+            -31, 4, -9, -8, -10, -10, -11, -11, -11, -11, -12, -11, -11,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,
+
+            /* Feature vec 2 */
+            -31, 4, -9, -9, -10, -10, -11, -11, -11, -11, -12, -12, -12,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,
+
+            /* Feature vec 3 */
+            -31, 4, -9, -9, -10, -10, -11, -11, -11, -11, -11, -12, -12,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,
+
+            /* Feature vec 4 : this should have valid delta 1 and delta 2 */
+            -31, 4, -9, -9, -10, -10, -11, -11, -11, -11, -11, -12, -12,
+            -38, -29, -9, 1, -2, -7, -8, -8, -12, -16, -14, -5, 5,
+            -68, -50, -13, 5, 0, -9, -9, -8, -13, -20, -19, -3, 15,
+
+            /* Feature vec 5 : this should have valid delta 1 and delta 2 */
+            -31, 4, -9, -8, -10, -10, -11, -11, -11, -11, -11, -12, -12,
+            -62, -45, -11, 5, 0, -8, -9, -8, -12, -19, -17, -3, 13,
+            -27, -22, -13, -9, -11, -12, -12, -11, -11, -13, -13, -10, -6,
+
+            /* Feature vec 6 */
+            -31, 4, -9, -8, -10, -10, -11, -11, -11, -11, -12, -11, -11,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,
+
+            /* Feature vec 7 */
+            -32, 4, -9, -8, -10, -10, -11, -11, -11, -12, -12, -11, -11,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,
+
+            /* Feature vec 8 */
+            -32, 4, -9, -8, -10, -10, -11, -11, -11, -12, -12, -11, -11,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10,
+
+            /* Feature vec 9 */
+            -31, 4, -9, -8, -10, -10, -11, -11, -11, -11, -12, -11, -11,
+            -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11, -11,
+            -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10, -10
+    };
+
+    /* Check that the elements have been calculated correctly */
+    for (uint32_t j = 0; j < numMfccVectors; ++j)
+    {
+        for (uint32_t i = 0; i < numMfccFeats * 3; ++i)
+        {
+            size_t tensorIdx = (j * numMfccFeats * 3) + i;
+            CHECK(static_cast<int16_t>(outputBuffer.at(tensorIdx) == static_cast<int16_t>(expectedResult[j][i])));
+        }
+    }
+}