OLA Algorithm and test shell.

BUG=16011
TEST=none
Review URL: http://codereview.chromium.org/151120

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@19994 0039d316-1c4b-4281-b951-d872f2087c98

4 files changed, 452 insertions, 0 deletions

diff --git a/media/filters/audio_renderer_algorithm_ola.cc b/media/filters/audio_renderer_algorithm_ola.cc
new file mode 100644
index 0000000..9a235de
--- /dev/null
+++ b/media/filters/audio_renderer_algorithm_ola.cc
@@ -0,0 +1,212 @@
+// Copyright (c) 2009 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "media/filters/audio_renderer_algorithm_ola.h"
+
+#include <cmath>
+
+#include "media/base/buffers.h"
+#include "media/base/data_buffer.h"
+
+namespace media {
+
+// Default window size in bytes.
+// TODO(kylep): base the window size in seconds, not bytes.
+const size_t kDefaultWindowSize = 4096;
+
+AudioRendererAlgorithmOLA::AudioRendererAlgorithmOLA()
+    : data_offset_(0),
+      input_step_(0),
+      output_step_(0) {
+}
+
+AudioRendererAlgorithmOLA::~AudioRendererAlgorithmOLA() {
+}
+
+size_t AudioRendererAlgorithmOLA::FillBuffer(DataBuffer* buffer_out) {
+  if (IsInputFinished())
+    return 0;
+  if (playback_rate() == 0.0f)
+    return 0;
+
+  // Grab info from |buffer_out| and handle the simple case of normal playback.
+  size_t dest_remaining = buffer_out->GetDataSize();
+  uint8* dest = buffer_out->GetWritableData(dest_remaining);
+  size_t dest_written = 0;
+  if (playback_rate() == 1.0f) {
+    dest_written = CopyInput(dest, dest_remaining);
+    AdvanceInput(dest_written);
+    return dest_written;
+  }
+
+  // For other playback rates, OLA with crossfade!
+  // TODO(kylep): Limit the rates to reasonable values. We may want to do this
+  // on the UI side or in set_playback_rate().
+  while (dest_remaining >= output_step_ + crossfade_size_) {
+    // Copy bulk of data to output (including some to crossfade to the next
+    // copy), then add to our running sum of written data and subtract from
+    // our tally of remaing requested.
+    size_t copied = CopyInput(dest, output_step_ + crossfade_size_);
+    dest_written += copied;
+    dest_remaining -= copied;
+
+    // Advance pointers for crossfade.
+    dest += output_step_;
+    AdvanceInput(input_step_);
+
+    // Prepare intermediate buffer.
+    size_t crossfade_size;
+    scoped_array<uint8> src(new uint8[crossfade_size_]);
+    crossfade_size = CopyInput(src.get(), crossfade_size_);
+
+    // Calculate number of samples to crossfade, then do so.
+    int samples = static_cast<int>(crossfade_size / sample_bytes()
+        / channels());
+    switch (sample_bytes()) {
+      case 4:
+        Crossfade(samples,
+            reinterpret_cast<const int32*>(src.get()),
+            reinterpret_cast<int32*>(dest));
+        break;
+      case 2:
+        Crossfade(samples,
+            reinterpret_cast<const int16*>(src.get()),
+            reinterpret_cast<int16*>(dest));
+        break;
+      case 1:
+        Crossfade(samples, src.get(), dest);
+        break;
+      default:
+        NOTREACHED() << "Unsupported audio bit depth sent to OLA algorithm";
+    }
+
+    // Advance pointers again.
+    AdvanceInput(crossfade_size_);
+    dest += crossfade_size_;
+  }
+  return dest_written;
+}
+
+void AudioRendererAlgorithmOLA::FlushBuffers() {
+  AudioRendererAlgorithmBase::FlushBuffers();
+  saved_buf_ = NULL;
+}
+
+void AudioRendererAlgorithmOLA::set_playback_rate(float new_rate) {
+  AudioRendererAlgorithmBase::set_playback_rate(new_rate);
+
+  // Adjusting step sizes to accomodate requested playback rate.
+  if (playback_rate() > 1.0f) {
+    input_step_ = kDefaultWindowSize;
+    output_step_ = static_cast<size_t>(ceil(
+        static_cast<float>(kDefaultWindowSize / playback_rate())));
+  } else {
+    input_step_ = static_cast<size_t>(ceil(
+        static_cast<float>(kDefaultWindowSize * playback_rate())));
+    output_step_ = kDefaultWindowSize;
+  }
+  AlignToSampleBoundary(&input_step_);
+  AlignToSampleBoundary(&output_step_);
+
+  // Calculate length for crossfading.
+  crossfade_size_ = kDefaultWindowSize / 10;
+  AlignToSampleBoundary(&crossfade_size_);
+
+  // To keep true to playback rate, modify the steps.
+  input_step_ -= crossfade_size_;
+  output_step_ -= crossfade_size_;
+}
+
+void AudioRendererAlgorithmOLA::AdvanceInput(size_t bytes) {
+  if (IsInputFinished())
+    return;
+
+  DCHECK(saved_buf_) << "Did you forget to call CopyInput()?";
+
+  // Calculate number of usable bytes in |saved_buf_|.
+  size_t saved_buf_remaining = saved_buf_->GetDataSize() - data_offset_;
+
+  // If there is enough data in |saved_buf_| to advance into it, do so.
+  // Otherwise, advance into the queue.
+  if (saved_buf_remaining > bytes) {
+    data_offset_ += bytes;
+  } else {
+    if (!IsQueueEmpty()) {
+      saved_buf_ = FrontQueue();
+      PopFrontQueue();
+    } else {
+      saved_buf_ = NULL;
+    }
+    // TODO(kylep): Make this function loop to eliminate the DCHECK.
+    DCHECK_GE(bytes, saved_buf_remaining);
+
+    data_offset_ = bytes - saved_buf_remaining;
+  }
+}
+
+void AudioRendererAlgorithmOLA::AlignToSampleBoundary(size_t* value) {
+  (*value) -= ((*value) % (channels() * sample_bytes()));
+}
+
+// TODO(kylep): Make this function loop to satisfy requests better.
+size_t AudioRendererAlgorithmOLA::CopyInput(uint8* dest, size_t length) {
+  if (IsInputFinished())
+    return 0;
+
+  // Lazy initialization.
+  if (!saved_buf_) {
+    saved_buf_ = FrontQueue();
+    PopFrontQueue();
+  }
+
+  size_t dest_written = 0;
+  size_t data_length = saved_buf_->GetDataSize() - data_offset_;
+
+  // Prevent writing past end of the buffer.
+  if (data_length > length)
+    data_length = length;
+  memcpy(dest, saved_buf_->GetData() + data_offset_, data_length);
+
+  dest += data_length;
+  length -= data_length;
+  dest_written += data_length;
+
+  if (length > 0) {
+    // We should have enough data in the next buffer so long as the
+    // queue is not empty.
+    if (IsQueueEmpty())
+      return dest_written;
+    DCHECK_LE(length, FrontQueue()->GetDataSize());
+
+    memcpy(dest, FrontQueue()->GetData(), length);
+    dest_written += length;
+  }
+
+  return dest_written;
+}
+
+template <class Type>
+void AudioRendererAlgorithmOLA::Crossfade(int samples,
+                                          const Type* src,
+                                          Type* dest) {
+  Type* dest_end = dest + samples * channels();
+  const Type* src_end = src + samples * channels();
+  for (int i = 0; i < samples; ++i) {
+    double x_ratio = static_cast<double>(i) / static_cast<double>(samples);
+    for (int j = 0; j < channels(); ++j) {
+      DCHECK(dest < dest_end);
+      DCHECK(src < src_end);
+      (*dest) = static_cast<Type>((*dest) * (1.0 - x_ratio) +
+                                  (*src) * x_ratio);
+      ++src;
+      ++dest;
+    }
+  }
+}
+
+bool AudioRendererAlgorithmOLA::IsInputFinished() {
+  return !saved_buf_ && IsQueueEmpty();
+}
+
+}  // namespace media
diff --git a/media/filters/audio_renderer_algorithm_ola.h b/media/filters/audio_renderer_algorithm_ola.h
new file mode 100644
index 0000000..c46ba134
--- /dev/null
+++ b/media/filters/audio_renderer_algorithm_ola.h
@@ -0,0 +1,75 @@
+// Copyright (c) 2009 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// AudioRendererAlgorithmOLA [ARAO] is the pitch-preservation implementation of
+// AudioRendererAlgorithmBase [ARAB]. For speeds greater than 1.0f, FillBuffer()
+// consumes more input data than output data requested and crossfades
+// samples to fill |buffer_out|. For speeds less than 1.0f, FillBuffer()
+// consumers less input data than output data requested and draws overlapping
+// samples from the input data to fill |buffer_out|. As ARAB is thread-unsafe,
+// so is ARAO.
+
+#ifndef MEDIA_FILTERS_AUDIO_RENDERER_ALGORITHM_OLA_H_
+#define MEDIA_FILTERS_AUDIO_RENDERER_ALGORITHM_OLA_H_
+
+#include "media/filters/audio_renderer_algorithm_base.h"
+
+namespace media {
+
+class DataBuffer;
+
+class AudioRendererAlgorithmOLA : public AudioRendererAlgorithmBase {
+ public:
+  AudioRendererAlgorithmOLA();
+  virtual ~AudioRendererAlgorithmOLA();
+
+  // AudioRendererAlgorithmBase implementation
+  virtual size_t FillBuffer(DataBuffer* buffer_out);
+
+  virtual void FlushBuffers();
+
+  virtual void set_playback_rate(float new_rate);
+
+ private:
+  // Advances our input position by |bytes| bytes.
+  void AdvanceInput(size_t bytes);
+
+  // Aligns |value| to a channel and sample boundary.
+  void AlignToSampleBoundary(size_t* value);
+
+  // Copies |length| bytes from our buffers to |dest|. Returns how many bytes
+  // of data were copied.
+  size_t CopyInput(uint8* dest, size_t length);
+
+  // Crossfades |samples| samples of |dest| with the data in |src|. Assumes
+  // there is room in |dest| and enough data in |src|. Type is the datatype
+  // of a data point in the waveform (i.e. uint8, int16, int32, etc). Also,
+  // sizeof(one sample) == sizeof(Type) * channels.
+  template <class Type>
+  void Crossfade(int samples, const Type* src, Type* dest);
+
+  // Returns true if |saved_buf_| is NULL and the queue is empty.
+  bool IsInputFinished();
+
+  // Stores the front element of the queue so we can peek at the second one.
+  scoped_refptr<Buffer> saved_buf_;
+
+  // Remembers the amount of remaining audio data for our |saved_buf_|.
+  size_t data_offset_;
+
+  // Members for ease of calculation in FillBuffer(). These members are
+  // based on |playback_rate_|, but are stored seperately so they don't
+  // have to be recalculated on every call to FillBuffer().
+  size_t input_step_;
+  size_t output_step_;
+
+  // Length for crossfade in bytes.
+  size_t crossfade_size_;
+
+  DISALLOW_COPY_AND_ASSIGN(AudioRendererAlgorithmOLA);
+};
+
+}  // namespace media
+
+#endif  // MEDIA_FILTERS_AUDIO_RENDERER_ALGORITHM_OLA_H_
diff --git a/media/media.gyp b/media/media.gyp
index 39751e2..1b20982 100644
--- a/media/media.gyp
+++ b/media/media.gyp
@@ -76,6 +76,8 @@
         'filters/audio_renderer_algorithm_base.h',
         'filters/audio_renderer_algorithm_default.cc',
         'filters/audio_renderer_algorithm_default.h',
+        'filters/audio_renderer_algorithm_ola.cc',
+        'filters/audio_renderer_algorithm_ola.h',
         'filters/audio_renderer_base.cc',
         'filters/audio_renderer_base.h',
         'filters/audio_renderer_impl.cc',
@@ -192,6 +194,16 @@
         'bench/bench.cc',
       ],
     },
+    {
+      'target_name': 'wav_ola_test',
+      'type': 'executable',
+      'dependencies': [
+        'media',
+      ],
+      'sources': [
+        'tools/wav_ola_test.cc'
+      ],
+    },
   ],
   'conditions': [
     ['OS=="win"', {
diff --git a/media/tools/wav_ola_test.cc b/media/tools/wav_ola_test.cc
new file mode 100644
index 0000000..59c6970
--- /dev/null
+++ b/media/tools/wav_ola_test.cc
@@ -0,0 +1,153 @@
+// Copyright (c) 2009 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// This application is a test for AudioRendererAlgorithmOLA. It reads in a
+// specified wav file (so far only 8, 16 and 32 bit are supported) and uses
+// ARAO to scale the playback by a specified rate. Then it outputs the result
+// to the specified location. Command line calls should be as follows:
+//
+//    wav_ola_test RATE INFILE OUTFILE
+
+#include <iostream>
+#include <string>
+
+#include "base/command_line.h"
+#include "base/file_util.h"
+#include "base/ref_counted.h"
+#include "base/string_util.h"
+#include "media/base/data_buffer.h"
+#include "media/filters/audio_renderer_algorithm_ola.h"
+
+using file_util::ScopedFILE;
+using media::AudioRendererAlgorithmOLA;
+using media::DataBuffer;
+
+const size_t kDefaultWindowSize = 4096;
+
+struct WavHeader {
+  int32 riff;
+  int32 chunk_size;
+  char unused0[8];
+  int32 subchunk1_size;
+  int16 audio_format;
+  int16 channels;
+  int32 sample_rate;
+  char unused1[6];
+  int16 bit_rate;
+  char unused2[4];
+  int32 subchunk2_size;
+};
+
+// Dummy class to feed data to OLA algorithm. Necessary to create callback.
+class Dummy {
+ public:
+  Dummy(FILE* in, AudioRendererAlgorithmOLA* ola)
+      : input_(in),
+        ola_(ola) {
+  }
+
+  void ReadDataForAlg() {
+    scoped_refptr<DataBuffer> b(new DataBuffer());
+    uint8* buf = b->GetWritableData(kDefaultWindowSize);
+    if (fread(buf, 1, kDefaultWindowSize, input_) > 0) {
+      ola_->EnqueueBuffer(b.get());
+    }
+  }
+
+ private:
+  FILE* input_;
+  AudioRendererAlgorithmOLA* ola_;
+
+  DISALLOW_COPY_AND_ASSIGN(Dummy);
+};
+
+int main(int argc, const char** argv) {
+  AudioRendererAlgorithmOLA ola;
+  CommandLine::Init(argc, argv);
+  const CommandLine* cmd_line = CommandLine::ForCurrentProcess();
+
+  std::vector<std::wstring> filenames(cmd_line->GetLooseValues());
+  if (filenames.empty()) {
+    std::cerr << "Usage: alg_test RATE INFILE OUTFILE\n"
+              << std::endl;
+    return 1;
+  }
+
+  // Retrieve command line options.
+  std::string in_path(WideToUTF8(filenames[1]));
+  std::string out_path(WideToUTF8(filenames[2]));
+  double playback_rate = 0.0;
+
+  // Determine speed of rerecord.
+  if (!StringToDouble(WideToUTF8(filenames[0]), &playback_rate))
+    playback_rate = 0.0;
+
+  // Open input file.
+  ScopedFILE input(file_util::OpenFile(in_path.c_str(), "rb"));
+  if (!(input.get())) {
+    LOG(ERROR) << "could not open input";
+    return 1;
+  }
+
+  // Open output file.
+  ScopedFILE output(file_util::OpenFile(out_path.c_str(), "wb"));
+  if (!(output.get())) {
+    LOG(ERROR) << "could not open output";
+    return 1;
+  }
+
+  // Read in header.
+  WavHeader wav;
+  if (fread(&wav, sizeof(wav), 1, input.get()) < 1) {
+    LOG(ERROR) << "could not read WAV header";
+    return 1;
+  }
+
+  // Instantiate dummy class and callback to feed data to |ola|.
+  Dummy guy(input.get(), &ola);
+  AudioRendererAlgorithmOLA::RequestReadCallback* cb =
+      NewCallback(&guy, &Dummy::ReadDataForAlg);
+  ola.Initialize(wav.channels,
+                 wav.bit_rate,
+                 static_cast<float>(playback_rate),
+                 cb);
+
+  // Print out input format.
+  std::cout << in_path << "\n"
+            << "Channels: " << wav.channels << "\n"
+            << "Sample Rate: " << wav.sample_rate << "\n"
+            << "Bit Rate: " << wav.bit_rate << "\n"
+            << "\n"
+            << "Scaling audio by " << playback_rate << "x..." << std::endl;
+
+  // Write the header back out again.
+  if (fwrite(&wav, sizeof(wav), 1, output.get()) < 1) {
+    LOG(ERROR) << "could not write WAV header";
+    return 1;
+  }
+
+  // Create buffer to be filled by |ola|.
+  scoped_refptr<DataBuffer> buffer(new DataBuffer());
+  uint8* buf = buffer->GetWritableData(kDefaultWindowSize);
+
+  // Keep track of bytes written to disk and bytes copied to |b|.
+  size_t bytes_written = 0;
+  size_t bytes;
+  while ((bytes = ola.FillBuffer(buffer.get())) > 0) {
+    if (fwrite(buf, 1, bytes, output.get()) != bytes) {
+      LOG(ERROR) << "could not write data after " << bytes_written;
+    } else {
+      bytes_written += bytes;
+    }
+  }
+
+  // Seek back to the beginning of our output file and update the header.
+  wav.chunk_size = 36 + bytes_written;
+  wav.subchunk1_size = 16;
+  wav.subchunk2_size = bytes_written;
+  fseek(output.get(), 0, SEEK_SET);
+  fwrite(&wav, sizeof(wav), 1, output.get());
+
+  return 0;
+}