AudioDecoderSpeex::AudioDecoderSpeex()
: _speex_dec_state(speex_decoder_init(&speex_wb_mode))
{
if (!_speex_dec_state) {
throw MediaException(_("AudioDecoderSpeex: state initialization failed."));
}
speex_bits_init(&_speex_bits);
speex_decoder_ctl(_speex_dec_state, SPEEX_GET_FRAME_SIZE, &_speex_framesize);
#ifdef RESAMPLING_SPEEX
int err = 0;
_resampler = speex_resampler_init(1, 16000, 44100,
SPEEX_RESAMPLER_QUALITY_DEFAULT, &err);
if (err != RESAMPLER_ERR_SUCCESS) {
throw MediaException(_("AudioDecoderSpeex: initialization failed."));
}
spx_uint32_t num = 0, den = 0;
speex_resampler_get_ratio (_resampler, &num, &den);
assert(num && den);
boost::rational<boost::uint32_t> numsamples(den, num);
numsamples *= _speex_framesize * 2 /* convert to stereo */;
_target_frame_size = boost::rational_cast<boost::uint32_t>(numsamples);
#endif
}
// Resamples input data to an output buffer, according to |mBufferSampleRate| and
// the playbackRate/detune.
// The number of frames consumed/produced depends on the amount of space
// remaining in both the input and output buffer, and the playback rate (that
// is, the ratio between the output samplerate and the input samplerate).
void CopyFromInputBufferWithResampling(AudioBlock* aOutput,
uint32_t aChannels,
uint32_t* aOffsetWithinBlock,
uint32_t aAvailableInOutput,
StreamTime* aCurrentPosition,
uint32_t aBufferMax)
{
if (*aOffsetWithinBlock == 0) {
aOutput->AllocateChannels(aChannels);
}
SpeexResamplerState* resampler = mResampler;
MOZ_ASSERT(aChannels > 0);
if (mBufferPosition < aBufferMax) {
uint32_t availableInInputBuffer = aBufferMax - mBufferPosition;
uint32_t ratioNum, ratioDen;
speex_resampler_get_ratio(resampler, &ratioNum, &ratioDen);
// Limit the number of input samples copied and possibly
// format-converted for resampling by estimating how many will be used.
// This may be a little small if still filling the resampler with
// initial data, but we'll get called again and it will work out.
uint32_t inputLimit = aAvailableInOutput * ratioNum / ratioDen + 10;
if (!BegunResampling()) {
// First time the resampler is used.
uint32_t inputLatency = speex_resampler_get_input_latency(resampler);
inputLimit += inputLatency;
// If starting after mStart, then play from the beginning of the
// buffer, but correct for input latency. If starting before mStart,
// then align the resampler so that the time corresponding to the
// first input sample is mStart.
int64_t skipFracNum = static_cast<int64_t>(inputLatency) * ratioDen;
double leadTicks = mStart - *aCurrentPosition;
if (leadTicks > 0.0) {
// Round to nearest output subsample supported by the resampler at
// these rates.
int64_t leadSubsamples = leadTicks * ratioNum + 0.5;
MOZ_ASSERT(leadSubsamples <= skipFracNum,
"mBeginProcessing is wrong?");
skipFracNum -= leadSubsamples;
}
speex_resampler_set_skip_frac_num(resampler,
std::min<int64_t>(skipFracNum, UINT32_MAX));
mBeginProcessing = -STREAM_TIME_MAX;
}
inputLimit = std::min(inputLimit, availableInInputBuffer);
MOZ_ASSERT(mBuffer.mVolume == 1.0f);
for (uint32_t i = 0; true; ) {
uint32_t inSamples = inputLimit;
uint32_t outSamples = aAvailableInOutput;
float* outputData =
aOutput->ChannelFloatsForWrite(i) + *aOffsetWithinBlock;
if (mBuffer.mBufferFormat == AUDIO_FORMAT_FLOAT32) {
const float* inputData =
mBuffer.ChannelData<float>()[i] + mBufferPosition;
WebAudioUtils::SpeexResamplerProcess(resampler, i,
inputData, &inSamples,
outputData, &outSamples);
} else {
MOZ_ASSERT(mBuffer.mBufferFormat == AUDIO_FORMAT_S16);
const int16_t* inputData =
mBuffer.ChannelData<int16_t>()[i] + mBufferPosition;
WebAudioUtils::SpeexResamplerProcess(resampler, i,
inputData, &inSamples,
outputData, &outSamples);
}
if (++i == aChannels) {
mBufferPosition += inSamples;
MOZ_ASSERT(mBufferPosition <= mBufferEnd || mLoop);
*aOffsetWithinBlock += outSamples;
*aCurrentPosition += outSamples;
if (inSamples == availableInInputBuffer && !mLoop) {
// We'll feed in enough zeros to empty out the resampler's memory.
// This handles the output latency as well as capturing the low
// pass effects of the resample filter.
mRemainingResamplerTail =
2 * speex_resampler_get_input_latency(resampler) - 1;
}
return;
}
}
} else {
for (uint32_t i = 0; true; ) {
uint32_t inSamples = mRemainingResamplerTail;
uint32_t outSamples = aAvailableInOutput;
float* outputData =
aOutput->ChannelFloatsForWrite(i) + *aOffsetWithinBlock;
// AudioDataValue* for aIn selects the function that does not try to
// copy and format-convert input data.
WebAudioUtils::SpeexResamplerProcess(resampler, i,
static_cast<AudioDataValue*>(nullptr), &inSamples,
outputData, &outSamples);
if (++i == aChannels) {
MOZ_ASSERT(inSamples <= mRemainingResamplerTail);
mRemainingResamplerTail -= inSamples;
*aOffsetWithinBlock += outSamples;
*aCurrentPosition += outSamples;
break;
}
}
}
}
