123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319 |
- // 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>
- #include <memory>
- #include <vector>
- #include "base/bind.h"
- #include "base/callback.h"
- #include "base/logging.h"
- #include "base/numerics/safe_math.h"
- #include "base/time/time.h"
- #include "media/base/audio_bus.h"
- #include "media/base/audio_sample_types.h"
- #include "media/ffmpeg/ffmpeg_common.h"
- #include "media/ffmpeg/ffmpeg_decoding_loop.h"
- namespace media {
- // AAC(M4A) decoding specific constants.
- static const int kAACPrimingFrameCount = 2112;
- static const int kAACRemainderFrameCount = 519;
- AudioFileReader::AudioFileReader(FFmpegURLProtocol* protocol)
- : stream_index_(0),
- protocol_(protocol),
- audio_codec_(AudioCodec::kUnknown),
- channels_(0),
- sample_rate_(0),
- av_sample_format_(0) {}
- AudioFileReader::~AudioFileReader() {
- Close();
- }
- bool AudioFileReader::Open() {
- return OpenDemuxer() && OpenDecoder();
- }
- bool AudioFileReader::OpenDemuxer() {
- glue_ = std::make_unique<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;
- }
- const int result = avformat_find_stream_info(format_context, NULL);
- if (result < 0) {
- DLOG(WARNING)
- << "AudioFileReader::Open() : error in avformat_find_stream_info()";
- return false;
- }
- // Calling avformat_find_stream_info can uncover new streams. We wait till now
- // to find the first audio stream, if any.
- codec_context_.reset();
- bool found_stream = false;
- for (size_t i = 0; i < format_context->nb_streams; ++i) {
- if (format_context->streams[i]->codecpar->codec_type ==
- AVMEDIA_TYPE_AUDIO) {
- stream_index_ = i;
- found_stream = true;
- break;
- }
- }
- if (!found_stream)
- return false;
- // Get the codec context.
- codec_context_ =
- AVStreamToAVCodecContext(format_context->streams[stream_index_]);
- if (!codec_context_)
- return false;
- DCHECK_EQ(codec_context_->codec_type, AVMEDIA_TYPE_AUDIO);
- return true;
- }
- bool AudioFileReader::OpenDecoder() {
- const AVCodec* codec = avcodec_find_decoder(codec_context_->codec_id);
- if (codec) {
- // MP3 decodes to S16P which we don't support, tell it to use S16 instead.
- if (codec_context_->sample_fmt == AV_SAMPLE_FMT_S16P)
- codec_context_->request_sample_fmt = AV_SAMPLE_FMT_S16;
- const int result = avcodec_open2(codec_context_.get(), codec, nullptr);
- if (result < 0) {
- DLOG(WARNING) << "AudioFileReader::Open() : could not open codec -"
- << " result: " << result;
- return false;
- }
- // Ensure avcodec_open2() respected our format request.
- if (codec_context_->sample_fmt == AV_SAMPLE_FMT_S16P) {
- DLOG(ERROR) << "AudioFileReader::Open() : unable to configure a"
- << " supported sample format - "
- << codec_context_->sample_fmt;
- return false;
- }
- } else {
- DLOG(WARNING) << "AudioFileReader::Open() : could not find codec.";
- return false;
- }
- // Verify the channel layout is supported by Chrome. Acts as a sanity check
- // against invalid files. See http://crbug.com/171962
- if (ChannelLayoutToChromeChannelLayout(
- codec_context_->ch_layout.u.mask,
- codec_context_->ch_layout.nb_channels) ==
- CHANNEL_LAYOUT_UNSUPPORTED) {
- return false;
- }
- // Store initial values to guard against midstream configuration changes.
- channels_ = codec_context_->ch_layout.nb_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(
- std::vector<std::unique_ptr<AudioBus>>* decoded_audio_packets,
- int packets_to_read) {
- DCHECK(glue_ && codec_context_)
- << "AudioFileReader::Read() : reader is not opened!";
- FFmpegDecodingLoop decode_loop(codec_context_.get());
- int total_frames = 0;
- auto frame_ready_cb =
- base::BindRepeating(&AudioFileReader::OnNewFrame, base::Unretained(this),
- &total_frames, decoded_audio_packets);
- AVPacket packet;
- int packets_read = 0;
- while (packets_read++ < packets_to_read && ReadPacket(&packet)) {
- const auto status = decode_loop.DecodePacket(&packet, frame_ready_cb);
- av_packet_unref(&packet);
- if (status != FFmpegDecodingLoop::DecodeStatus::kOkay)
- break;
- }
- return total_frames;
- }
- 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_ == AudioCodec::kAAC) {
- // 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
- // especially noticeable for short files (< 10ms) resulting in silence
- // throughout the decoded buffer. Thus we add the priming frames and the
- // remainder frames to the estimation.
- // (See: crbug.com/513178)
- estimated_duration_us += ceil(
- 1000000.0 *
- static_cast<double>(kAACPrimingFrameCount + kAACRemainderFrameCount) /
- sample_rate());
- } else {
- // Add one microsecond to avoid rounding-down errors which can occur when
- // |duration| has been calculated from an exact number of sample-frames.
- // One microsecond is much less than the time of a single sample-frame
- // at any real-world sample-rate.
- estimated_duration_us += 1;
- }
- return ConvertFromTimeBase(av_time_base, estimated_duration_us.ValueOrDie());
- }
- int AudioFileReader::GetNumberOfFrames() const {
- return base::ClampCeil(GetDuration().InSecondsF() * sample_rate());
- }
- bool AudioFileReader::OpenDemuxerForTesting() {
- return OpenDemuxer();
- }
- bool AudioFileReader::ReadPacketForTesting(AVPacket* output_packet) {
- return ReadPacket(output_packet);
- }
- bool AudioFileReader::ReadPacket(AVPacket* output_packet) {
- while (av_read_frame(glue_->format_context(), output_packet) >= 0) {
- // Skip packets from other streams.
- if (output_packet->stream_index != stream_index_) {
- av_packet_unref(output_packet);
- continue;
- }
- return true;
- }
- return false;
- }
- bool AudioFileReader::OnNewFrame(
- int* total_frames,
- std::vector<std::unique_ptr<AudioBus>>* decoded_audio_packets,
- AVFrame* frame) {
- int frames_read = frame->nb_samples;
- if (frames_read < 0)
- return false;
- const int channels = frame->ch_layout.nb_channels;
- if (frame->sample_rate != sample_rate_ || channels != channels_ ||
- frame->format != av_sample_format_) {
- DLOG(ERROR) << "Unsupported midstream configuration change!"
- << " Sample Rate: " << frame->sample_rate << " vs "
- << sample_rate_ << ", Channels: " << channels << " vs "
- << channels_ << ", Sample Format: " << frame->format << " vs "
- << av_sample_format_;
- // This is an unrecoverable error, so bail out. We'll return
- // whatever we've decoded up to this point.
- return false;
- }
- // AAC decoding doesn't properly trim the last packet in a stream, so if we
- // have duration information, use it to set the correct length to avoid extra
- // silence from being output. In the case where we are also discarding some
- // portion of the packet (as indicated by a negative pts), we further want to
- // adjust the duration downward by however much exists before zero.
- if (audio_codec_ == AudioCodec::kAAC && frame->duration) {
- const base::TimeDelta pkt_duration = ConvertFromTimeBase(
- glue_->format_context()->streams[stream_index_]->time_base,
- frame->duration + std::min(static_cast<int64_t>(0), frame->pts));
- const base::TimeDelta frame_duration =
- base::Seconds(frames_read / static_cast<double>(sample_rate_));
- if (pkt_duration < frame_duration && pkt_duration.is_positive()) {
- const int new_frames_read =
- base::ClampFloor(frames_read * (pkt_duration / frame_duration));
- DVLOG(2) << "Shrinking AAC frame from " << frames_read << " to "
- << new_frames_read << " based on packet duration.";
- frames_read = new_frames_read;
- // The above process may delete the entire packet.
- if (!frames_read)
- return true;
- }
- }
- // 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.
- decoded_audio_packets->emplace_back(AudioBus::Create(channels, frames_read));
- AudioBus* audio_bus = decoded_audio_packets->back().get();
- if (codec_context_->sample_fmt == AV_SAMPLE_FMT_FLT) {
- audio_bus->FromInterleaved<Float32SampleTypeTraits>(
- reinterpret_cast<float*>(frame->data[0]), frames_read);
- } else if (codec_context_->sample_fmt == AV_SAMPLE_FMT_FLTP) {
- for (int ch = 0; ch < audio_bus->channels(); ++ch) {
- memcpy(audio_bus->channel(ch), frame->extended_data[ch],
- sizeof(float) * frames_read);
- }
- } else {
- int bytes_per_sample = av_get_bytes_per_sample(codec_context_->sample_fmt);
- switch (bytes_per_sample) {
- case 1:
- audio_bus->FromInterleaved<UnsignedInt8SampleTypeTraits>(
- reinterpret_cast<const uint8_t*>(frame->data[0]), frames_read);
- break;
- case 2:
- audio_bus->FromInterleaved<SignedInt16SampleTypeTraits>(
- reinterpret_cast<const int16_t*>(frame->data[0]), frames_read);
- break;
- case 4:
- audio_bus->FromInterleaved<SignedInt32SampleTypeTraits>(
- reinterpret_cast<const int32_t*>(frame->data[0]), frames_read);
- break;
- default:
- NOTREACHED() << "Unsupported bytes per sample encountered: "
- << bytes_per_sample;
- audio_bus->ZeroFrames(frames_read);
- }
- }
- (*total_frames) += frames_read;
- return true;
- }
- bool AudioFileReader::SeekForTesting(base::TimeDelta seek_time) {
- // Use the AVStream's time_base, since |codec_context_| does not have
- // time_base populated until after OpenDecoder().
- return av_seek_frame(
- 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
|