如何将音频帧从 input.mp4 传递到 libavcodec 中的 output.mp4？

Question

I have a project that correctly opens a.mp4, extracts video frames, modifies them, and dumps the modified frames to an output.mp4. Everything works (mostly - I have a video timing bug that pops up at random, but I'll kill it) EXCEPT for the writing of audio. I don't want to modify the audio channel at all - I just want the audio from the input.mp4 to be passed, unaltered, to the output.mp4.

There's too much code here to provide a working example, largely because there's a lot of OpenGL and GLSL in there, but the most important part is where I advance a frame. This method is called in a loop, and if the frame was a video frame, the loop sends the image data to the rendering hardware, does a bunch of GL magic on it, then writes out a frame of video. If the frame was an audio frame, the loop does nothing, but the advance_frame() method is supposed to just dump that frame to the output mp4. I'm at a loss as to what libavcodec provides that will do this.

Note that here, I'm decoding the audio packets into frames, but that shouldn't be necessary. I'd rather work with packets and not burn the CPU time to do the decode at all. (I've tried it the other way, but this is what I wound up with when I tried to decode the data, then re-encode to create the output stream.) I just need a way to pass the packets from the input to the output.

bool MediaContainerMgr::advance_frame() {
    int ret; // Crappy naming, but I'm using ffmpeg's name for it.
    while (true) {
        ret = av_read_frame(m_format_context, m_packet);
        if (ret < 0) {
            // Do we actually need to unref the packet if it failed?
            av_packet_unref(m_packet);
            if (ret == AVERROR_EOF) {
                finalize_output();
                return false;
            }
            continue;
            //return false;
        }
        else {
            int response = decode_packet();
            if (response != 0) {
                continue;
            }
            // If this was an audio packet, the image renderer doesn't care about it - just push
            // the audio data to the output .mp4:
            if (m_packet->stream_index == m_audio_stream_index) {
                printf("m_packet->stream_index: %d\n", m_packet->stream_index);
                printf("  m_packet->pts: %lld\n", m_packet->pts);
                printf("  mpacket->size: %d\n", m_packet->size);
                // m_recording is true if we're writing a .mp4, as opposed to just letting OpenGL
                // display the frames onscreen.
                if (m_recording) {
                    int err = 0;
                    // I've tried everything I can find to try to push the audio frame to the
                    // output .mp4. This doesn't work, but neither do a half-dozen other libavcodec
                    // methods:
                    err = avcodec_send_frame(m_output_audio_codec_context, m_last_audio_frame);

                    if (err) {
                        printf("  encoding error: %d\n", err);
                    }
                }
            }
            av_packet_unref(m_packet);
            if (m_packet->stream_index == m_video_stream_index) {
                return true;
            }
        }
    }
}

The workhorse of advance_frame() is decode_packet() . All of this works perfectly for video data:

int MediaContainerMgr::decode_packet() {
    // Supply raw packet data as input to a decoder
    // https://ffmpeg.org/doxygen/trunk/group__lavc__decoding.html#ga58bc4bf1e0ac59e27362597e467efff3
    int             response;
    AVCodecContext* codec_context = nullptr;
    AVFrame*        frame         = nullptr;

    if (m_packet->stream_index == m_video_stream_index) {
        codec_context = m_video_input_codec_context;
        frame = m_last_video_frame;
    }
    if (m_packet->stream_index == m_audio_stream_index) {
        codec_context = m_audio_input_codec_context;
        frame = m_last_audio_frame;
    }

    if (codec_context == nullptr) {
        return -1;
    }

    response = avcodec_send_packet(codec_context, m_packet);
    if (response < 0) {
        char buf[256];
        av_strerror(response, buf, 256);
        printf("Error while receiving a frame from the decoder: %s\n", buf);
        return response;
    }

    // Return decoded output data (into a frame) from a decoder
    // https://ffmpeg.org/doxygen/trunk/group__lavc__decoding.html#ga11e6542c4e66d3028668788a1a74217c
    response = avcodec_receive_frame(codec_context, frame);
    if (response == AVERROR(EAGAIN) || response == AVERROR_EOF) {
        return response;
    } else if (response < 0) {
        char buf[256];
        av_strerror(response, buf, 256);
        printf("Error while receiving a frame from the decoder: %s\n", buf);
        return response;
    } else {
        printf(
            "Stream %d, Frame %d (type=%c, size=%d bytes), pts %lld, key_frame %d, [DTS %d]\n",
            m_packet->stream_index,
            codec_context->frame_number,
            av_get_picture_type_char(frame->pict_type),
            frame->pkt_size,
            frame->pts,
            frame->key_frame,
            frame->coded_picture_number
        );
    }
    return 0;
}

I can provide the setup for all the contexts, if necessary, but for brevity maybe we can get away with what av_dump_format(m_output_format_context, 0, filename, 1) displays:

Output #0, mp4, to 'D:\yodeling_monkey_nuggets.mp4':
  Metadata:
    encoder         : Lavf58.64.100
    Stream #0:0: Video: h264 (libx264) (avc1 / 0x31637661), yuv420p, 1920x1080, q=-1--1, 20305 kb/s, 29.97 fps, 30k tbn
    Stream #0:1: Audio: aac (mp4a / 0x6134706D), 44100 Hz, mono, fltp, 125 kb/s

Answer 1

To put audio AVPacket "as-is" to the output without decode-encode steps you should use av_write_frame function for such packets instead of avcodec_send_frame . Note that these functions use different contexts: AVFormatContext and AVCodecContext .

avcodec_send_frame supplies a raw video or audio frame to the encoder

av_write_frame passes the packet directly to the muxer