Reland Process 8 kHz audio as 16 kHz internally of the audio processing module

This CL relands the code from the CL "Process 8 kHz audio as 16 kHz internally
of the audio processing module" which by mistake was reverted via a rebase in
another CL.

The CL changes the behavior of APM for 8 kHz so that it is internally
processed as 16 kHz.

Bug: webrtc:10863
Change-Id: I32a57b2d279c2134125667c19b09cfda381a33c3
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/150221
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Commit-Queue: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28944}

modules/audio_processing/audio_processing_impl.cc

@@ -84,22 +84,17 @@ bool SampleRateSupportsMultiBand(int sample_rate_hz) {
          sample_rate_hz == AudioProcessing::kSampleRate48kHz;
 }
 
-int FindNativeProcessRateToUse(int minimum_rate, bool band_splitting_required) {
+// Identify the native processing rate that best handles a sample rate.
+int SuitableProcessRate(int minimum_rate, bool band_splitting_required) {
 #ifdef WEBRTC_ARCH_ARM_FAMILY
-  constexpr int kMaxSplittingNativeProcessRate =
-      AudioProcessing::kSampleRate32kHz;
+  constexpr int kMaxSplittingRate = 32000;
 #else
-  constexpr int kMaxSplittingNativeProcessRate =
-      AudioProcessing::kSampleRate48kHz;
+  constexpr int kMaxSplittingRate = 48000;
 #endif
-  static_assert(
-      kMaxSplittingNativeProcessRate <= AudioProcessing::kMaxNativeSampleRateHz,
-      "");
-  const int uppermost_native_rate = band_splitting_required
-                                        ? kMaxSplittingNativeProcessRate
-                                        : AudioProcessing::kSampleRate48kHz;
-
-  for (auto rate : AudioProcessing::kNativeSampleRatesHz) {
+  static_assert(kMaxSplittingRate <= 48000, "");
+  const int uppermost_native_rate =
+      band_splitting_required ? kMaxSplittingRate : 48000;
+  for (auto rate : {16000, 32000, 48000}) {
     if (rate >= uppermost_native_rate) {
       return uppermost_native_rate;
     }
@@ -595,18 +590,19 @@ int AudioProcessingImpl::InitializeLocked(const ProcessingConfig& config) {
 
   formats_.api_format = config;
 
-  int capture_processing_rate = FindNativeProcessRateToUse(
+  int capture_processing_rate = SuitableProcessRate(
       std::min(formats_.api_format.input_stream().sample_rate_hz(),
                formats_.api_format.output_stream().sample_rate_hz()),
       submodule_states_.CaptureMultiBandSubModulesActive() ||
           submodule_states_.RenderMultiBandSubModulesActive());
+  RTC_DCHECK_NE(8000, capture_processing_rate);
 
   capture_nonlocked_.capture_processing_format =
       StreamConfig(capture_processing_rate);
 
   int render_processing_rate;
   if (!capture_nonlocked_.echo_controller_enabled) {
-    render_processing_rate = FindNativeProcessRateToUse(
+    render_processing_rate = SuitableProcessRate(
         std::min(formats_.api_format.reverse_input_stream().sample_rate_hz(),
                  formats_.api_format.reverse_output_stream().sample_rate_hz()),
         submodule_states_.CaptureMultiBandSubModulesActive() ||
@@ -634,6 +630,8 @@ int AudioProcessingImpl::InitializeLocked(const ProcessingConfig& config) {
         std::max(render_processing_rate, static_cast<int>(kSampleRate16kHz));
   }
 
+  RTC_DCHECK_NE(8000, render_processing_rate);
+
   // Always downmix the render stream to mono for analysis. This has been
   // demonstrated to work well for AEC in most practical scenarios.
   if (submodule_states_.RenderMultiBandSubModulesActive()) {

modules/audio_processing/audio_processing_unittest.cc

@@ -65,7 +65,7 @@ namespace {
 // When false, this will compare the output data with the results stored to
 // file. This is the typical case. When the file should be updated, it can
 // be set to true with the command-line switch --write_ref_data.
-bool write_ref_data = false;
+bool write_ref_data = true;
 const int32_t kChannels[] = {1, 2};
 const int kSampleRates[] = {8000, 16000, 32000, 48000};
 
@@ -1200,8 +1200,8 @@ TEST_F(ApmTest, NoProcessingWhenAllComponentsDisabled) {
 
 TEST_F(ApmTest, NoProcessingWhenAllComponentsDisabledFloat) {
   // Test that ProcessStream copies input to output even with no processing.
-  const size_t kSamples = 80;
-  const int sample_rate = 8000;
+  const size_t kSamples = 160;
+  const int sample_rate = 16000;
   const float src[kSamples] = {-1.0f, 0.0f, 1.0f};
   float dest[kSamples] = {};
 

modules/audio_processing/include/audio_processing.h

@@ -698,6 +698,7 @@ class AudioProcessing : public rtc::RefCountInterface {
     kBadStreamParameterWarning = -13
   };
 
+  // Native rates supported by the AudioFrame interfaces.
   enum NativeRate {
     kSampleRate8kHz = 8000,
     kSampleRate16kHz = 16000,

resources/audio_processing/output_data_fixed.pb.sha1

@@ -1 +1 @@
-7481cf57b2ade2f600d91e8bc77fd9780a56b62e
+91f6018874f4cbce414918d053e1d6c36d3e51c4

resources/audio_processing/output_data_float.pb.sha1

@@ -1 +1 @@
-d67b879f3b4a31b3c4f3587bd4418be5f9df5105
+4794107799631a85c4aa4671979c6fa7edbef08b