Added a bitexactness test for the voice activity detector in the audio processing module.

webrtc/modules/audio_processing/voice_detection_unittest.cc

+/*
+ *  Copyright (c) 2015 The WebRTC project authors. All Rights Reserved.

[hlundin-webrtc, 2016/03/16 15:50:20]

2016

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#include <vector>
+
+#include "testing/gtest/include/gtest/gtest.h"
+#include "webrtc/base/array_view.h"
+#include "webrtc/base/random.h"
+#include "webrtc/modules/audio_coding/neteq/tools/audio_loop.h"
+#include "webrtc/modules/audio_processing/audio_buffer.h"
+#include "webrtc/modules/audio_processing/voice_detection_impl.h"
+#include "webrtc/modules/audio_processing/test/audio_buffer_tools.h"
+#include "webrtc/modules/audio_processing/test/bitexactness_tools.h"
+#include "webrtc/test/testsupport/fileutils.h"
+
+namespace webrtc {
+namespace {
+
+enum TestSignalLevels { kLow, kMedium, kHigh };
+
+::testing::AssertionResult AssertLikelihoodsNotEqual(

[hlundin-webrtc, 2016/03/16 15:50:20]

This is a lot of code to compare two enum (integer) values. Use EXPECT_EQ
instead.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+    const char* m_expr,
+    const char* n_expr,
+    const VoiceDetection::Likelihood& output,
+    const VoiceDetection::Likelihood& reference) {
+  // If the values are deemed not to be similar, return a report of the
+  // difference.
+  if (output != reference) {
+    // Lambda function that produces a string containing the likelihood name.
+    auto likelihood_description = [](VoiceDetection::Likelihood likelihood) {
+      switch (likelihood) {
+        case VoiceDetection::kVeryLowLikelihood:
+          return std::string("kVeryLowLikelihood");
+          break;
+        case VoiceDetection::kLowLikelihood:
+          return std::string("kLowLikelihood");
+          break;
+        case VoiceDetection::kModerateLikelihood:
+          return std::string("kModerateLikelihood");
+          break;
+        case VoiceDetection::kHighLikelihood:
+          return std::string("kHighLikelihood");
+          break;
+        default:
+          RTC_DCHECK(false);
+          return std::string("");
+      }
+    };
+
+    return ::testing::AssertionFailure()
+           << "Actual: " << likelihood_description(output) << std::endl
+           << "Expected: " << likelihood_description(reference) << std::endl;
+  }
+  return ::testing::AssertionSuccess();
+}
+
+// Process one frame of data and produce the output.
+void ProcessOneFrame(int sample_rate_hz,
+                     AudioBuffer* audio_buffer,
+                     VoiceDetectionImpl* voice_detection,
+                     int* frame_size_ms,
+                     bool* stream_has_voice,
+                     VoiceDetection::Likelihood* likelihood) {
+  if (sample_rate_hz > AudioProcessing::kSampleRate16kHz) {
+    audio_buffer->SplitIntoFrequencyBands();
+  }
+
+  voice_detection->ProcessCaptureAudio(audio_buffer);
+
+  if (sample_rate_hz > AudioProcessing::kSampleRate16kHz) {

[hlundin-webrtc, 2016/03/16 15:50:20]

What is the rationale for merging the bands again? Is this buffer used any more
after the voice detection is done?

[peah-webrtc, 2016/03/18 05:56:05]

No point at all in this case, I will remove that.

+    audio_buffer->MergeFrequencyBands();
+  }
+
+  *frame_size_ms = voice_detection->frame_size_ms();
+  *stream_has_voice = voice_detection->stream_has_voice();
+  *likelihood = voice_detection->likelihood();
+}
+
+// Forms a predefined random test vector.
+void ConstructTestVector(int samples_per_channel,
+                         int num_channels,
+                         test::AudioLoop* audio_loop,
+                         TestSignalLevels signal_level,
+                         std::vector<float>* testvector) {
+  testvector->resize(samples_per_channel * num_channels);
+  auto input_samples = audio_loop->GetNextBlock();
+
+  float signal_gain = 0.0f;

[hlundin-webrtc, 2016/03/16 15:50:20]

Consider:

float signal_gain = 0.1f;  // TestSignalLevels::kLow
if (signal_level == TestSignalLevels::kMedium)
  signal_gain = 0.5f;
else if (signal_level == TestSignalLevels::kHigh)
  signal_gain = 1.0f;

More compact, and if you make TestSignalLevels an enum class, you won't have to
worry about other values.

[peah-webrtc, 2016/03/18 05:56:05]

Thanks. This code is now removed.

+  switch (signal_level) {
+    case TestSignalLevels::kLow:
+      signal_gain = 0.1f;
+      break;
+    case TestSignalLevels::kMedium:
+      signal_gain = 0.5f;
+      break;
+    case TestSignalLevels::kHigh:
+      signal_gain = 1.0f;
+      break;
+    default:
+      RTC_DCHECK(false);
+  }
+
+  for (int k = 0; k < samples_per_channel; ++k) {
+    for (int j = 0; j < num_channels; ++j) {
+      (*testvector)[k * num_channels + j] =
+          signal_gain * input_samples[j * samples_per_channel + k] / 32768.0f;
+    }
+  }
+}
+
+void SetupComponent(int sample_rate_hz, VoiceDetectionImpl* voice_detection) {
+  voice_detection->Initialize(sample_rate_hz > 16000 ? 16000 : sample_rate_hz);
+  voice_detection->Enable(true);
+}
+
+std::string GetTestVectorFileName(int sample_rate_hz) {

[hlundin-webrtc, 2016/03/16 15:50:20]

You can get rid of this function if you follow my suggestion below.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+  switch (sample_rate_hz) {
+    case 8000:
+      // Use a the 16 kHz signal for the 8 kHz case as well. Acceptable since
+      // the test is only a bitexactness test.
+      return webrtc::test::ResourcePath("audio_coding/speech_mono_16kHz",
+                                        "pcm");
+    case 16000:
+      return webrtc::test::ResourcePath("audio_coding/speech_mono_16kHz",
+                                        "pcm");
+    case 32000:
+      return webrtc::test::ResourcePath("audio_coding/speech_mono_32_48kHz",
+                                        "pcm");
+    case 48000:
+      // Use a the 32 kHz signal for the 48 kHz case as well. Acceptable since
+      // the test is only a bitexactness test.
+      return webrtc::test::ResourcePath("audio_coding/speech_mono_32_48kHz",
+                                        "pcm");
+    default:
+      RTC_DCHECK(false);
+  }
+
+  return "";
+}
+
+// Processes a specified amount of frames, verifies the results and reports
+// any errors.
+void RunBitexactnessTest(int sample_rate_hz,
+                         int num_channels,
+                         int num_frames_to_process,
+                         TestSignalLevels signal_level,
+                         int frame_size_ms_reference,
+                         bool stream_has_voice_reference,
+                         VoiceDetection::Likelihood likelihood_reference) {
+  test::AudioLoop audio_loop;

[hlundin-webrtc, 2016/03/16 15:50:20]

I would actually recommend you use a ResampleInputAudioFile instead. It is a
specialization of InputAudioFile. The differences are that it reads directly
from file instead of from memory, but it supports resampling so you can use a
single input file for all rates. It also supports looping, just like AudioLoop.

Bonus: it has a DuplicateInterleaved utility.

[peah-webrtc, 2016/03/18 05:56:05]

Thanks for the suggestion!!! I found some Resource files for 8-48 kHz for
far-end and nearend so I will for now use those instead, using InputAudioFile.
But if they prove not to be sufficiently good, I'll instead go for
ResampleInputAudioFile.

+  int samples_per_channel = 80 * sample_rate_hz / 8000;
+  const StreamConfig stream_config(sample_rate_hz, num_channels, false);
+  AudioBuffer audio_buffer(
+      stream_config.num_frames(), stream_config.num_channels(),
+      stream_config.num_frames(), stream_config.num_channels(),
+      stream_config.num_frames());
+
+  std::string filename;

[hlundin-webrtc, 2016/03/16 15:50:20]

Not used.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+  bool success = audio_loop.Init(
+      GetTestVectorFileName(sample_rate_hz),
+      num_frames_to_process * samples_per_channel * num_channels,
+      samples_per_channel * num_channels);
+  RTC_DCHECK(success);

[hlundin-webrtc, 2016/03/16 15:50:20]

This is test code; you might as well CHECK things like this, since continuing
with the test without an input file makes no sense, release or debug alike.

[peah-webrtc, 2016/03/18 05:56:05]

Good point!!!
Removed this one but will change other DCHECK to CHECK within the tests.

+
+  rtc::CriticalSection crit;
+  VoiceDetectionImpl voice_detection(&crit);
+  SetupComponent(sample_rate_hz, &voice_detection);
+
+  std::vector<float> frame_input;
+  int frame_size_ms;
+  bool stream_has_voice;
+  VoiceDetection::Likelihood likelihood;
+  for (int frame_no = 0; frame_no < num_frames_to_process; ++frame_no) {
+    ConstructTestVector(samples_per_channel, num_channels, &audio_loop,
+                        signal_level, &frame_input);
+
+    test::CopyVectorToAudioBuffer(stream_config, frame_input, &audio_buffer);
+
+    ProcessOneFrame(sample_rate_hz, &audio_buffer, &voice_detection,
+                    &frame_size_ms, &stream_has_voice, &likelihood);
+  }
+
+  // Compare the outputs to the references.
+  EXPECT_PRED_FORMAT2(test::AssertIntegersNotEqual, frame_size_ms,

[hlundin-webrtc, 2016/03/16 15:50:20]

I think all of these should be changed to regular EXPECT_EQ.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+                      frame_size_ms_reference);
+  EXPECT_PRED_FORMAT2(test::AssertBoolsNotEqual, stream_has_voice,
+                      stream_has_voice_reference);
+  EXPECT_PRED_FORMAT2(AssertLikelihoodsNotEqual, likelihood,
+                      likelihood_reference);
+}
+
+const int kNumFramesToProcess = 1000;
+
+}  // namespace
+
+TEST(VoiceDetectionBitExactnessTest, Mono8kHzLow) {
+  const int kFrameSizeMsReference = 10;

[hlundin-webrtc, 2016/03/16 15:50:20]

kFrameSizeMsReference is always 10; define it once before the TESTs.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+  const bool kStreamHAsVoiceReference = true;

[hlundin-webrtc, 2016/03/16 15:50:20]

HAs -> Has

[hlundin-webrtc, 2016/03/16 15:50:20]

Define once before the TESTs.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

[peah-webrtc, 2016/03/18 05:56:06]

Done.

+  const VoiceDetection::Likelihood kLlikelihoodReference =

[hlundin-webrtc, 2016/03/16 15:50:20]

... and define it before the TESTs.

[hlundin-webrtc, 2016/03/16 15:50:20]

kLli -> kLi

[peah-webrtc, 2016/03/18 05:56:05]

Done.

[peah-webrtc, 2016/03/18 05:56:05]

Done.

+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(8000, 1, kNumFramesToProcess, TestSignalLevels::kLow,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono16kHzLow) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(16000, 1, kNumFramesToProcess, TestSignalLevels::kLow,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono32kHzLow) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(32000, 1, kNumFramesToProcess, TestSignalLevels::kLow,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono48kHzLow) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(48000, 1, kNumFramesToProcess, TestSignalLevels::kLow,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Stereo16kHzLow) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(16000, 2, kNumFramesToProcess, TestSignalLevels::kLow,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono8kHzMedium) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(8000, 1, kNumFramesToProcess, TestSignalLevels::kMedium,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono16kHzMedium) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(16000, 1, kNumFramesToProcess, TestSignalLevels::kMedium,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono32kHzMedium) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(32000, 1, kNumFramesToProcess, TestSignalLevels::kMedium,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono48kHzMedium) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(48000, 1, kNumFramesToProcess, TestSignalLevels::kMedium,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Stereo16kHzMedium) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(16000, 2, kNumFramesToProcess, TestSignalLevels::kMedium,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono8kHzHigh) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(8000, 1, kNumFramesToProcess, TestSignalLevels::kHigh,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono16kHzHigh) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(16000, 1, kNumFramesToProcess, TestSignalLevels::kHigh,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono32kHzHigh) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(32000, 1, kNumFramesToProcess, TestSignalLevels::kHigh,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Mono48kHzHigh) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(48000, 1, kNumFramesToProcess, TestSignalLevels::kHigh,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+TEST(VoiceDetectionBitExactnessTest, Stereo16kHzHigh) {
+  const int kFrameSizeMsReference = 10;
+  const bool kStreamHAsVoiceReference = true;
+  const VoiceDetection::Likelihood kLlikelihoodReference =
+      VoiceDetection::kLowLikelihood;
+
+  RunBitexactnessTest(16000, 2, kNumFramesToProcess, TestSignalLevels::kHigh,
+                      kFrameSizeMsReference, kStreamHAsVoiceReference,
+                      kLlikelihoodReference);
+}
+
+}  // namespace webrtc