overhaul of audio_provider_lavc.cpp. should fix the infamous skewing issue, tested and works on windows at least.

Originally committed to SVN as r2236.
This commit is contained in:
Karl Blomster 2008-07-04 12:04:10 +00:00
parent d01b4ec3e9
commit e26b9fe0d5

View file

@ -101,15 +101,15 @@ LAVCAudioProvider::LAVCAudioProvider(Aegisub::String _filename)
} }
if (audStream == -1) { if (audStream == -1) {
codecContext = NULL; codecContext = NULL;
throw _T("Could not find an audio stream"); throw _T("ffmpeg audio provider: Could not find an audio stream");
} }
AVCodec *codec = avcodec_find_decoder(codecContext->codec_id); AVCodec *codec = avcodec_find_decoder(codecContext->codec_id);
if (!codec) { if (!codec) {
codecContext = NULL; codecContext = NULL;
throw _T("Could not find a suitable audio decoder"); throw _T("ffmpeg audio provider: Could not find a suitable audio decoder");
} }
if (avcodec_open(codecContext, codec) < 0) if (avcodec_open(codecContext, codec) < 0)
throw _T("Failed to open audio decoder"); throw _T("ffmpeg audio provider: Failed to open audio decoder");
sample_rate = Options.AsInt(_T("Audio Sample Rate")); sample_rate = Options.AsInt(_T("Audio Sample Rate"));
if (!sample_rate) if (!sample_rate)
@ -124,7 +124,7 @@ LAVCAudioProvider::LAVCAudioProvider(Aegisub::String _filename)
if ((sample_rate != codecContext->sample_rate) || (codecContext->channels > 1)) { if ((sample_rate != codecContext->sample_rate) || (codecContext->channels > 1)) {
rsct = audio_resample_init(1, codecContext->channels, sample_rate, codecContext->sample_rate); rsct = audio_resample_init(1, codecContext->channels, sample_rate, codecContext->sample_rate);
if (!rsct) if (!rsct)
throw _T("Failed to initialize resampling"); throw _T("ffmpeg audio provider: Failed to initialize resampling");
resample_ratio = (float)sample_rate / (float)codecContext->sample_rate; resample_ratio = (float)sample_rate / (float)codecContext->sample_rate;
} }
@ -141,7 +141,7 @@ LAVCAudioProvider::LAVCAudioProvider(Aegisub::String _filename)
buffer = (int16_t *)malloc(AVCODEC_MAX_AUDIO_FRAME_SIZE); buffer = (int16_t *)malloc(AVCODEC_MAX_AUDIO_FRAME_SIZE);
if (!buffer) if (!buffer)
throw _T("Failed to allocate %d bytes for audio decoding buffer, out of memory?", AVCODEC_MAX_AUDIO_FRAME_SIZE); throw _T("ffmpeg audio provider: Failed to allocate audio decoding buffer, out of memory?");
leftover_samples = 0; leftover_samples = 0;
@ -172,6 +172,7 @@ void LAVCAudioProvider::Destroy()
void LAVCAudioProvider::GetAudio(void *buf, int64_t start, int64_t count) void LAVCAudioProvider::GetAudio(void *buf, int64_t start, int64_t count)
{ {
int16_t *_buf = (int16_t *)buf; int16_t *_buf = (int16_t *)buf;
int64_t samples_to_decode = num_samples - start; /* samples left to the end of the stream */ int64_t samples_to_decode = num_samples - start; /* samples left to the end of the stream */
if (count < samples_to_decode) /* haven't reached the end yet, so just decode the requested number of samples */ if (count < samples_to_decode) /* haven't reached the end yet, so just decode the requested number of samples */
samples_to_decode = count; samples_to_decode = count;
@ -182,49 +183,88 @@ void LAVCAudioProvider::GetAudio(void *buf, int64_t start, int64_t count)
we have enough to fill the request */ we have enough to fill the request */
memset(_buf + samples_to_decode, 0, (count - samples_to_decode) * 2); memset(_buf + samples_to_decode, 0, (count - samples_to_decode) * 2);
/* do we have leftover samples from last time we were called? */
if (leftover_samples > 0) {
/* put them in the output buffer */
samples_to_decode -= leftover_samples;
for (std::vector<int16_t>::iterator i = overshoot_buffer.begin(); i != overshoot_buffer.end(); i++) {
*(_buf++) = *i;
}
/* none left */
leftover_samples = 0;
overshoot_buffer.clear();
}
AVPacket packet; AVPacket packet;
while (samples_to_decode > 0 && av_read_frame(lavcfile->fctx, &packet) >= 0) { while (samples_to_decode > 0 && av_read_frame(lavcfile->fctx, &packet) >= 0) {
/* we're not dealing with video packets in this here provider */ /* we're not dealing with video packets in this here provider */
if (packet.stream_index == audStream) { if (packet.stream_index == audStream) {
int size = packet.size; int size = packet.size;
uint8_t *data = packet.data;
while (size > 0) { while (size > 0) {
int temp_output_buffer_size = AVCODEC_MAX_AUDIO_FRAME_SIZE; /* see constructor, it malloc()'s buffer to this */ int temp_output_buffer_size = AVCODEC_MAX_AUDIO_FRAME_SIZE; /* see constructor, it malloc()'s buffer to this */
int retval, decoded_samples; int retval, decoded_bytes, decoded_samples;
retval = avcodec_decode_audio2(codecContext, buffer, &temp_output_buffer_size, data, size); retval = avcodec_decode_audio2(codecContext, buffer, &temp_output_buffer_size, packet.data, size);
if (retval <= 0) if (retval <= 0)
throw _T("Failed to decode audio"); throw _T("ffmpeg audio provider: failed to decode audio");
/* decoding succeeded but the output buffer is empty, go to next packet */
if (temp_output_buffer_size == 0)
continue;
decoded_samples = temp_output_buffer_size / 2; /* 2 bytes per sample */ decoded_bytes = temp_output_buffer_size;
size -= retval; decoded_samples = decoded_bytes / 2; /* 2 bytes per sample */
data += retval; size -= decoded_bytes;
/* do we need to resample? */ /* do we need to resample? */
if (rsct) { if (rsct) {
/* do the actual resampling */ /* allocate some memory to save the resampled data in */
decoded_samples = audio_resample(rsct, _buf, buffer, decoded_samples / codecContext->channels); int16_t *temp_output_buffer = (int16_t *)malloc(AVCODEC_MAX_AUDIO_FRAME_SIZE);
if (!temp_output_buffer)
throw _T("ffmpeg audio provider: Failed to allocate audio resampling buffer, out of memory?");
/* do the actual resampling */
decoded_samples = audio_resample(rsct, temp_output_buffer, buffer, decoded_samples / codecContext->channels);
/* did we end up with more samples than we were asked for? */
if (decoded_samples > samples_to_decode) { if (decoded_samples > samples_to_decode) {
wxLogMessage(wxString::Format(_T("Warning: decoder output more samples than requested, audio skew highly likely! (Wanted %d, got %d)"), (int)samples_to_decode, decoded_samples)); /* in that case, count them */
leftover_samples = decoded_samples - samples_to_decode;
/* and put them aside for later */
overshoot_buffer = std::vector<int16_t>(&temp_output_buffer[samples_to_decode+1], &temp_output_buffer[decoded_samples+1]);
/* output the other samples that didn't overflow */
memcpy(_buf, temp_output_buffer, samples_to_decode * 2);
_buf += samples_to_decode;
} else {
memcpy(_buf, temp_output_buffer, decoded_samples * 2);
_buf += decoded_samples;
} }
} else { free(temp_output_buffer);
/* no resampling needed, just copy to the buffer, but first make noise if we got an overflow */ } else { /* no resampling needed */
if (decoded_samples > samples_to_decode) /* overflow? (as above) */
wxLogMessage(wxString::Format(_T("Warning: decoder output more samples than requested, audio skew highly likely! (Wanted %d, got %d)"), (int)samples_to_decode, decoded_samples)); if (decoded_samples > samples_to_decode) {
/* count sheep^H^H^H^H^Hsamples */
memcpy(_buf, buffer, temp_output_buffer_size); leftover_samples = decoded_samples - samples_to_decode;
/* and put them aside for later (mm, lamb chops) */
overshoot_buffer = std::vector<int16_t>(&buffer[samples_to_decode+1], &buffer[decoded_samples+1]);
/* output the other samples that didn't overflow */
memcpy(_buf, buffer, samples_to_decode * 2);
_buf += samples_to_decode;
} else {
/* just do a straight copy to buffer */
memcpy(_buf, buffer, decoded_bytes);
_buf += decoded_samples;
}
} }
_buf += decoded_samples;
samples_to_decode -= decoded_samples; samples_to_decode -= decoded_samples;
} }
} }
av_free_packet(&packet); av_free_packet(&packet);
} }
} }
#endif #endif