Allow more than 2 input channels in AudioProcessing.

The number of output channels is constrained to be equal to either 1 or the
number of input channels.

An earlier version of this commit caused a crash on AEC dump.

TBR=aluebs@webrtc.org,pbos@webrtc.org

Review URL: https://codereview.webrtc.org/1248393003 .

Cr-Commit-Position: refs/heads/master@{#9626}

4 files changed, 207 insertions, 34 deletions

diff --git a/webrtc/common_audio/audio_util.cc b/webrtc/common_audio/audio_util.cc
index 2047295cb9..be67c48f60 100644
--- a/webrtc/common_audio/audio_util.cc
+++ b/webrtc/common_audio/audio_util.cc
@@ -39,4 +39,13 @@ void FloatS16ToFloat(const float* src, size_t size, float* dest) {
     dest[i] = FloatS16ToFloat(src[i]);
 }
 
+template <>
+void DownmixInterleavedToMono<int16_t>(const int16_t* interleaved,
+                                       int num_multichannel_frames,
+                                       int num_channels,
+                                       int16_t* deinterleaved) {
+  DownmixInterleavedToMonoImpl<int16_t, int32_t>(
+      interleaved, num_multichannel_frames, num_channels, deinterleaved);
+}
+
 }  // namespace webrtc
diff --git a/webrtc/common_audio/audio_util_unittest.cc b/webrtc/common_audio/audio_util_unittest.cc
index 2cdf53813c..3ac3911445 100644
--- a/webrtc/common_audio/audio_util_unittest.cc
+++ b/webrtc/common_audio/audio_util_unittest.cc
@@ -8,11 +8,15 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include "testing/gmock/include/gmock/gmock.h"
 #include "testing/gtest/include/gtest/gtest.h"
 #include "webrtc/common_audio/include/audio_util.h"
 #include "webrtc/typedefs.h"
 
 namespace webrtc {
+namespace {
+
+using ::testing::ElementsAreArray;
 
 void ExpectArraysEq(const int16_t* ref, const int16_t* test, int length) {
   for (int i = 0; i < length; ++i) {
@@ -28,11 +32,17 @@ void ExpectArraysEq(const float* ref, const float* test, int length) {
 
 TEST(AudioUtilTest, FloatToS16) {
   const int kSize = 9;
-  const float kInput[kSize] = {
-      0.f, 0.4f / 32767.f, 0.6f / 32767.f, -0.4f / 32768.f, -0.6f / 32768.f,
-      1.f, -1.f, 1.1f, -1.1f};
-  const int16_t kReference[kSize] = {
-    0, 0, 1, 0, -1, 32767, -32768, 32767, -32768};
+  const float kInput[kSize] = {0.f,
+                               0.4f / 32767.f,
+                               0.6f / 32767.f,
+                               -0.4f / 32768.f,
+                               -0.6f / 32768.f,
+                               1.f,
+                               -1.f,
+                               1.1f,
+                               -1.1f};
+  const int16_t kReference[kSize] = {0,     0,      1,     0,     -1,
+                                     32767, -32768, 32767, -32768};
   int16_t output[kSize];
   FloatToS16(kInput, kSize, output);
   ExpectArraysEq(kReference, output, kSize);
@@ -50,8 +60,8 @@ TEST(AudioUtilTest, S16ToFloat) {
 
 TEST(AudioUtilTest, FloatS16ToS16) {
   const int kSize = 7;
-  const float kInput[kSize] = {
-      0.f, 0.4f, 0.5f, -0.4f, -0.5f, 32768.f, -32769.f};
+  const float kInput[kSize] = {0.f,   0.4f,    0.5f,    -0.4f,
+                               -0.5f, 32768.f, -32769.f};
   const int16_t kReference[kSize] = {0, 0, 1, 0, -1, 32767, -32768};
   int16_t output[kSize];
   FloatS16ToS16(kInput, kSize, output);
@@ -60,11 +70,17 @@ TEST(AudioUtilTest, FloatS16ToS16) {
 
 TEST(AudioUtilTest, FloatToFloatS16) {
   const int kSize = 9;
-  const float kInput[kSize] = {
-      0.f, 0.4f / 32767.f, 0.6f / 32767.f, -0.4f / 32768.f, -0.6f / 32768.f,
-      1.f, -1.f, 1.1f, -1.1f};
-  const float kReference[kSize] = {
-    0.f, 0.4f, 0.6f, -0.4f, -0.6f, 32767.f, -32768.f, 36043.7f, -36044.8f};
+  const float kInput[kSize] = {0.f,
+                               0.4f / 32767.f,
+                               0.6f / 32767.f,
+                               -0.4f / 32768.f,
+                               -0.6f / 32768.f,
+                               1.f,
+                               -1.f,
+                               1.1f,
+                               -1.1f};
+  const float kReference[kSize] = {0.f,     0.4f,     0.6f,     -0.4f,    -0.6f,
+                                   32767.f, -32768.f, 36043.7f, -36044.8f};
   float output[kSize];
   FloatToFloatS16(kInput, kSize, output);
   ExpectArraysEq(kReference, output, kSize);
@@ -72,11 +88,17 @@ TEST(AudioUtilTest, FloatToFloatS16) {
 
 TEST(AudioUtilTest, FloatS16ToFloat) {
   const int kSize = 9;
-  const float kInput[kSize] = {
-    0.f, 0.4f, 0.6f, -0.4f, -0.6f, 32767.f, -32768.f, 36043.7f, -36044.8f};
-  const float kReference[kSize] = {
-      0.f, 0.4f / 32767.f, 0.6f / 32767.f, -0.4f / 32768.f, -0.6f / 32768.f,
-      1.f, -1.f, 1.1f, -1.1f};
+  const float kInput[kSize] = {0.f,     0.4f,     0.6f,     -0.4f,    -0.6f,
+                               32767.f, -32768.f, 36043.7f, -36044.8f};
+  const float kReference[kSize] = {0.f,
+                                   0.4f / 32767.f,
+                                   0.6f / 32767.f,
+                                   -0.4f / 32768.f,
+                                   -0.6f / 32768.f,
+                                   1.f,
+                                   -1.f,
+                                   1.1f,
+                                   -1.1f};
   float output[kSize];
   FloatS16ToFloat(kInput, kSize, output);
   ExpectArraysEq(kReference, output, kSize);
@@ -114,4 +136,96 @@ TEST(AudioUtilTest, InterleavingMonoIsIdentical) {
   ExpectArraysEq(mono, interleaved, kSamplesPerChannel);
 }
 
+TEST(AudioUtilTest, DownmixInterleavedToMono) {
+  {
+    const int kNumFrames = 4;
+    const int kNumChannels = 1;
+    const int16_t interleaved[kNumChannels * kNumFrames] = {1, 2, -1, -3};
+    int16_t deinterleaved[kNumFrames];
+
+    DownmixInterleavedToMono(interleaved, kNumFrames, kNumChannels,
+                             deinterleaved);
+
+    EXPECT_THAT(deinterleaved, ElementsAreArray(interleaved));
+  }
+  {
+    const int kNumFrames = 2;
+    const int kNumChannels = 2;
+    const int16_t interleaved[kNumChannels * kNumFrames] = {10, 20, -10, -30};
+    int16_t deinterleaved[kNumFrames];
+
+    DownmixInterleavedToMono(interleaved, kNumFrames, kNumChannels,
+                             deinterleaved);
+    const int16_t expected[kNumFrames] = {15, -20};
+
+    EXPECT_THAT(deinterleaved, ElementsAreArray(expected));
+  }
+  {
+    const int kNumFrames = 3;
+    const int kNumChannels = 3;
+    const int16_t interleaved[kNumChannels * kNumFrames] = {
+        30000, 30000, 24001, -5, -10, -20, -30000, -30999, -30000};
+    int16_t deinterleaved[kNumFrames];
+
+    DownmixInterleavedToMono(interleaved, kNumFrames, kNumChannels,
+                             deinterleaved);
+    const int16_t expected[kNumFrames] = {28000, -11, -30333};
+
+    EXPECT_THAT(deinterleaved, ElementsAreArray(expected));
+  }
+}
+
+TEST(AudioUtilTest, DownmixToMonoTest) {
+  {
+    const int kNumFrames = 4;
+    const int kNumChannels = 1;
+    const float input_data[kNumChannels][kNumFrames] = {{1.f, 2.f, -1.f, -3.f}};
+    const float* input[kNumChannels];
+    for (int i = 0; i < kNumChannels; ++i) {
+      input[i] = input_data[i];
+    }
+
+    float downmixed[kNumFrames];
+
+    DownmixToMono<float, float>(input, kNumFrames, kNumChannels, downmixed);
+
+    EXPECT_THAT(downmixed, ElementsAreArray(input_data[0]));
+  }
+  {
+    const int kNumFrames = 3;
+    const int kNumChannels = 2;
+    const float input_data[kNumChannels][kNumFrames] = {{1.f, 2.f, -1.f},
+                                                        {3.f, 0.f, 1.f}};
+    const float* input[kNumChannels];
+    for (int i = 0; i < kNumChannels; ++i) {
+      input[i] = input_data[i];
+    }
+
+    float downmixed[kNumFrames];
+    const float expected[kNumFrames] = {2.f, 1.f, 0.f};
+
+    DownmixToMono<float, float>(input, kNumFrames, kNumChannels, downmixed);
+
+    EXPECT_THAT(downmixed, ElementsAreArray(expected));
+  }
+  {
+    const int kNumFrames = 3;
+    const int kNumChannels = 3;
+    const int16_t input_data[kNumChannels][kNumFrames] = {
+        {30000, -5, -30000}, {30000, -10, -30999}, {24001, -20, -30000}};
+    const int16_t* input[kNumChannels];
+    for (int i = 0; i < kNumChannels; ++i) {
+      input[i] = input_data[i];
+    }
+
+    int16_t downmixed[kNumFrames];
+    const int16_t expected[kNumFrames] = {28000, -11, -30333};
+
+    DownmixToMono<int16_t, int32_t>(input, kNumFrames, kNumChannels, downmixed);
+
+    EXPECT_THAT(downmixed, ElementsAreArray(expected));
+  }
+}
+
+}  // namespace
 }  // namespace webrtc
diff --git a/webrtc/common_audio/include/audio_util.h b/webrtc/common_audio/include/audio_util.h
index 8262649145..b217c683fd 100644
--- a/webrtc/common_audio/include/audio_util.h
+++ b/webrtc/common_audio/include/audio_util.h
@@ -12,7 +12,9 @@
 #define WEBRTC_COMMON_AUDIO_INCLUDE_AUDIO_UTIL_H_
 
 #include <limits>
+#include <cstring>
 
+#include "webrtc/base/checks.h"
 #include "webrtc/base/scoped_ptr.h"
 #include "webrtc/typedefs.h"
 
@@ -26,10 +28,10 @@ typedef std::numeric_limits<int16_t> limits_int16;
 // FloatS16: float   [-32768.0, 32767.0]
 static inline int16_t FloatToS16(float v) {
   if (v > 0)
-    return v >= 1 ? limits_int16::max() :
-                    static_cast<int16_t>(v * limits_int16::max() + 0.5f);
-  return v <= -1 ? limits_int16::min() :
-                   static_cast<int16_t>(-v * limits_int16::min() - 0.5f);
+    return v >= 1 ? limits_int16::max()
+                  : static_cast<int16_t>(v * limits_int16::max() + 0.5f);
+  return v <= -1 ? limits_int16::min()
+                 : static_cast<int16_t>(-v * limits_int16::min() - 0.5f);
 }
 
 static inline float S16ToFloat(int16_t v) {
@@ -42,10 +44,9 @@ static inline int16_t FloatS16ToS16(float v) {
   static const float kMaxRound = limits_int16::max() - 0.5f;
   static const float kMinRound = limits_int16::min() + 0.5f;
   if (v > 0)
-    return v >= kMaxRound ? limits_int16::max() :
-                            static_cast<int16_t>(v + 0.5f);
-  return v <= kMinRound ? limits_int16::min() :
-                          static_cast<int16_t>(v - 0.5f);
+    return v >= kMaxRound ? limits_int16::max()
+                          : static_cast<int16_t>(v + 0.5f);
+  return v <= kMinRound ? limits_int16::min() : static_cast<int16_t>(v - 0.5f);
 }
 
 static inline float FloatToFloatS16(float v) {
@@ -69,8 +70,10 @@ void FloatS16ToFloat(const float* src, size_t size, float* dest);
 // |deinterleaved| buffers (|num_channel| buffers with |samples_per_channel|
 // per buffer).
 template <typename T>
-void Deinterleave(const T* interleaved, int samples_per_channel,
-                  int num_channels, T* const* deinterleaved) {
+void Deinterleave(const T* interleaved,
+                  int samples_per_channel,
+                  int num_channels,
+                  T* const* deinterleaved) {
   for (int i = 0; i < num_channels; ++i) {
     T* channel = deinterleaved[i];
     int interleaved_idx = i;
@@ -85,8 +88,10 @@ void Deinterleave(const T* interleaved, int samples_per_channel,
 // |interleaved|. There must be sufficient space allocated in |interleaved|
 // (|samples_per_channel| * |num_channels|).
 template <typename T>
-void Interleave(const T* const* deinterleaved, int samples_per_channel,
-                int num_channels, T* interleaved) {
+void Interleave(const T* const* deinterleaved,
+                int samples_per_channel,
+                int num_channels,
+                T* interleaved) {
   for (int i = 0; i < num_channels; ++i) {
     const T* channel = deinterleaved[i];
     int interleaved_idx = i;
@@ -97,6 +102,56 @@ void Interleave(const T* const* deinterleaved, int samples_per_channel,
   }
 }
 
+template <typename T, typename Intermediate>
+void DownmixToMono(const T* const* input_channels,
+                   int num_frames,
+                   int num_channels,
+                   T* out) {
+  for (int i = 0; i < num_frames; ++i) {
+    Intermediate value = input_channels[0][i];
+    for (int j = 1; j < num_channels; ++j) {
+      value += input_channels[j][i];
+    }
+    out[i] = value / num_channels;
+  }
+}
+
+// Downmixes an interleaved multichannel signal to a single channel by averaging
+// all channels.
+template <typename T, typename Intermediate>
+void DownmixInterleavedToMonoImpl(const T* interleaved,
+                                  int num_frames,
+                                  int num_channels,
+                                  T* deinterleaved) {
+  DCHECK_GT(num_channels, 0);
+  DCHECK_GT(num_frames, 0);
+
+  const T* const end = interleaved + num_frames * num_channels;
+
+  while (interleaved < end) {
+    const T* const frame_end = interleaved + num_channels;
+
+    Intermediate value = *interleaved++;
+    while (interleaved < frame_end) {
+      value += *interleaved++;
+    }
+
+    *deinterleaved++ = value / num_channels;
+  }
+}
+
+template <typename T>
+void DownmixInterleavedToMono(const T* interleaved,
+                              int num_frames,
+                              int num_channels,
+                              T* deinterleaved);
+
+template <>
+void DownmixInterleavedToMono<int16_t>(const int16_t* interleaved,
+                                       int num_frames,
+                                       int num_channels,
+                                       int16_t* deinterleaved);
+
 }  // namespace webrtc
 
 #endif  // WEBRTC_COMMON_AUDIO_INCLUDE_AUDIO_UTIL_H_
diff --git a/webrtc/common_audio/wav_file.cc b/webrtc/common_audio/wav_file.cc
index 995043461a..a0c792c54a 100644
--- a/webrtc/common_audio/wav_file.cc
+++ b/webrtc/common_audio/wav_file.cc
@@ -123,11 +123,6 @@ void WavWriter::WriteSamples(const int16_t* samples, size_t num_samples) {
   num_samples_ += static_cast<uint32_t>(written);
   CHECK(written <= std::numeric_limits<uint32_t>::max() ||
         num_samples_ >= written);  // detect uint32_t overflow
-  CHECK(CheckWavParameters(num_channels_,
-                           sample_rate_,
-                           kWavFormat,
-                           kBytesPerSample,
-                           num_samples_));
 }
 
 void WavWriter::WriteSamples(const float* samples, size_t num_samples) {