Decode entire in-memory file for WebAudio

media/filters/audio_file_reader.cc

 // Copyright (c) 2012 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_file_reader.h"
 
 #include <stddef.h>
 
 #include <cmath>
 
@@ skipping 15 matching lines @@
       audio_codec_(kUnknownAudioCodec),
       channels_(0),
       sample_rate_(0),
       av_sample_format_(0) {}
 
 AudioFileReader::~AudioFileReader() {
   Close();
 }
 
 bool AudioFileReader::Open() {
-  if (!OpenDemuxer())
-    return false;
-  if (!OpenDecoder())
-    return false;
-
-  // If the duration is unknown, fail out; this API can not work with streams of
-  // unknown duration currently.
-  return glue_->format_context()->duration != AV_NOPTS_VALUE;
+  return OpenDemuxer() && OpenDecoder();
 }
 
 bool AudioFileReader::OpenDemuxer() {
   glue_.reset(new FFmpegGlue(protocol_));
   AVFormatContext* format_context = glue_->format_context();
 
   // Open FFmpeg AVFormatContext.
   if (!glue_->OpenContext()) {
     DLOG(WARNING) << "AudioFileReader::Open() : error in avformat_open_input()";
     return false;
@@ skipping 66 matching lines @@
   }
 
   // Store initial values to guard against midstream configuration changes.
   channels_ = codec_context_->channels;
   audio_codec_ = CodecIDToAudioCodec(codec_context_->codec_id);
   sample_rate_ = codec_context_->sample_rate;
   av_sample_format_ = codec_context_->sample_fmt;
   return true;
 }
 
+bool AudioFileReader::HasKnownDuration() const {
+  return glue_->format_context()->duration != AV_NOPTS_VALUE;
+}
+
 void AudioFileReader::Close() {
   codec_context_.reset();
   glue_.reset();
 }
 
 int AudioFileReader::Read(AudioBus* audio_bus) {
   DCHECK(glue_.get() && codec_context_) <<
       "AudioFileReader::Read() : reader is not opened!";
 
   DCHECK_EQ(audio_bus->channels(), channels());
@@ skipping 103 matching lines @@
 
   // Zero any remaining frames.
   audio_bus->ZeroFramesPartial(
       current_frame, audio_bus->frames() - current_frame);
 
   // Returns the actual number of sample-frames decoded.
   // Ideally this represents the "true" exact length of the file.
   return current_frame;
 }
 
+std::unique_ptr<AudioBus> AudioFileReader::StreamingRead() {

[DaleCurtis, 2017/01/25 22:42:25]

Needs refactor, so we don't have two copies of the decoding loop.

+  DCHECK(glue_.get() && codec_context_)
+      << "AudioFileReader::StreamingRead() : reader is not opened!";
+  size_t bytes_per_sample = av_get_bytes_per_sample(codec_context_->sample_fmt);
+
+  // Holds decoded audio.
+  std::unique_ptr<AVFrame, ScopedPtrAVFreeFrame> av_frame(av_frame_alloc());
+
+  // Holds a vector of AudioBus' containing each decoded packet.
+  std::vector<std::unique_ptr<AudioBus>> decodedAudioPackets;
+
+  AVPacket packet;
+  int current_frame = 0;
+  size_t total_frames = 0;
+  bool continue_decoding = true;
+
+  while (continue_decoding && ReadPacket(&packet)) {
+    // Make a shallow copy of packet so we can slide packet.data as frames are
+    // decoded from the packet; otherwise av_packet_unref() will corrupt memory.
+    AVPacket packet_temp = packet;
+    do {
+      // Reset frame to default values.
+      av_frame_unref(av_frame.get());
+
+      int frame_decoded = 0;
+      int result = avcodec_decode_audio4(codec_context_.get(), av_frame.get(),
+                                         &frame_decoded, &packet_temp);
+
+      if (result < 0) {
+        DLOG(WARNING)
+            << "AudioFileReader::Read() : error in avcodec_decode_audio4() -"
+            << result;
+        break;
+      }
+
+      // Update packet size and data pointer in case we need to call the decoder
+      // with the remaining bytes from this packet.
+      packet_temp.size -= result;
+      packet_temp.data += result;
+
+      if (!frame_decoded)
+        continue;
+
+      // Determine the number of sample-frames we just decoded.  Check overflow.
+      int frames_read = av_frame->nb_samples;
+      total_frames += frames_read;
+
+      if (frames_read < 0) {
+        continue_decoding = false;
+        break;
+      }
+
+#ifdef CHROMIUM_NO_AVFRAME_CHANNELS
+      int channels =
+          av_get_channel_layout_nb_channels(av_frame->channel_layout);
+#else
+      int channels = av_frame->channels;
+#endif
+      if (av_frame->sample_rate != sample_rate_ || channels != channels_ ||
+          av_frame->format != av_sample_format_) {
+        DLOG(ERROR) << "Unsupported midstream configuration change!"
+                    << " Sample Rate: " << av_frame->sample_rate << " vs "
+                    << sample_rate_ << ", Channels: " << channels << " vs "
+                    << channels_ << ", Sample Format: " << av_frame->format
+                    << " vs " << av_sample_format_;
+
+        // This is an unrecoverable error, so bail out.
+        continue_decoding = false;
+        break;
+      }
+
+      // Deinterleave each channel and convert to 32bit floating-point with
+      // nominal range -1.0 -> +1.0.  If the output is already in float planar
+      // format, just copy it into the AudioBus.
+      std::unique_ptr<AudioBus> audio_bus =
+          AudioBus::Create(channels, frames_read);
+
+      if (codec_context_->sample_fmt == AV_SAMPLE_FMT_FLT) {

[DaleCurtis, 2017/01/25 22:42:25]

I believe there's an AudioBus::FromInterleaved method for this now.

+        float* decoded_audio_data = reinterpret_cast<float*>(av_frame->data[0]);
+        int channels = audio_bus->channels();
+        for (int ch = 0; ch < channels; ++ch) {
+          float* bus_data = audio_bus->channel(ch) + current_frame;
+          for (int i = 0, offset = ch; i < frames_read;
+               ++i, offset += channels) {
+            bus_data[i] = decoded_audio_data[offset];
+          }
+        }
+      } else if (codec_context_->sample_fmt == AV_SAMPLE_FMT_FLTP) {
+        for (int ch = 0; ch < audio_bus->channels(); ++ch) {
+          memcpy(audio_bus->channel(ch) + current_frame,
+                 av_frame->extended_data[ch], sizeof(float) * frames_read);
+        }
+      } else {
+        audio_bus->FromInterleavedPartial(av_frame->data[0], current_frame,
+                                          frames_read, bytes_per_sample);
+      }
+
+      decodedAudioPackets.push_back(std::move(audio_bus));
+      // current_frame += frames_read;
+    } while (packet_temp.size > 0);
+    av_packet_unref(&packet);
+  }
+
+  // Concatenate all the decoded chunks into one big bus.
+  std::unique_ptr<AudioBus> outputBus =
+      AudioBus::Create(channels_, total_frames);
+
+  int dest_start_frame = 0;
+  for (size_t k = 0; k < decodedAudioPackets.size(); ++k) {
+    int frame_count = decodedAudioPackets[k]->frames();
+    decodedAudioPackets[k]->CopyPartialFramesTo(
+        0, frame_count, dest_start_frame, outputBus.get());
+    dest_start_frame += frame_count;
+  }
+
+  return outputBus;
+}
+
 base::TimeDelta AudioFileReader::GetDuration() const {
   const AVRational av_time_base = {1, AV_TIME_BASE};
 
   DCHECK_NE(glue_->format_context()->duration, AV_NOPTS_VALUE);
   base::CheckedNumeric<int64_t> estimated_duration_us =
       glue_->format_context()->duration;
 
   if (audio_codec_ == kCodecAAC) {
     // For certain AAC-encoded files, FFMPEG's estimated frame count might not
     // be sufficient to capture the entire audio content that we want. This is
@@ skipping 47 matching lines @@
              glue_->format_context(), stream_index_,
              ConvertToTimeBase(GetAVStreamForTesting()->time_base, seek_time),
              AVSEEK_FLAG_BACKWARD) >= 0;
 }
 
 const AVStream* AudioFileReader::GetAVStreamForTesting() const {
   return glue_->format_context()->streams[stream_index_];
 }
 
 }  // namespace media