static av_cold void opus_decode_flush(AVCodecContext *ctx)
{
OpusContext *c = ctx->priv_data;
int i;
for (i = 0; i < c->nb_streams; i++) {
OpusStreamContext *s = &c->streams[i];
memset(&s->packet, 0, sizeof(s->packet));
s->delayed_samples = 0;
if (s->celt_delay)
av_audio_fifo_drain(s->celt_delay, av_audio_fifo_size(s->celt_delay));
#if CONFIG_SWRESAMPLE
swr_close(s->swr);
#elif CONFIG_AVRESAMPLE
avresample_close(s->avr);
#endif
av_audio_fifo_drain(c->sync_buffers[i], av_audio_fifo_size(c->sync_buffers[i]));
ff_silk_flush(s->silk);
ff_celt_flush(s->celt);
}
}
static av_cold void opus_decode_flush(AVCodecContext *ctx)
{
OpusContext *c = ctx->priv_data;
int i;
for (i = 0; i < c->nb_streams; i++) {
OpusStreamContext *s = &c->streams[i];
memset(&s->packet, 0, sizeof(s->packet));
s->delayed_samples = 0;
if (s->celt_delay)
av_audio_fifo_drain(s->celt_delay, av_audio_fifo_size(s->celt_delay));
swr_close(s->swr);
ff_silk_flush(s->silk);
ff_celt_flush(s->celt);
}
}
static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size)
{
int samples = s->packet.frame_duration;
int redundancy = 0;
int redundancy_size, redundancy_pos;
int ret, i, consumed;
int delayed_samples = s->delayed_samples;
ret = opus_rc_init(&s->rc, data, size);
if (ret < 0)
return ret;
/* decode the silk frame */
if (s->packet.mode == OPUS_MODE_SILK || s->packet.mode == OPUS_MODE_HYBRID) {
if (!swr_is_initialized(s->swr)) {
ret = opus_init_resample(s);
if (ret < 0)
return ret;
}
samples = ff_silk_decode_superframe(s->silk, &s->rc, s->silk_output,
FFMIN(s->packet.bandwidth, OPUS_BANDWIDTH_WIDEBAND),
s->packet.stereo + 1,
silk_frame_duration_ms[s->packet.config]);
if (samples < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error decoding a SILK frame.\n");
return samples;
}
samples = swr_convert(s->swr,
(uint8_t**)s->out, s->packet.frame_duration,
(const uint8_t**)s->silk_output, samples);
if (samples < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error resampling SILK data.\n");
return samples;
}
av_assert2((samples & 7) == 0);
s->delayed_samples += s->packet.frame_duration - samples;
} else
ff_silk_flush(s->silk);
// decode redundancy information
consumed = opus_rc_tell(&s->rc);
if (s->packet.mode == OPUS_MODE_HYBRID && consumed + 37 <= size * 8)
redundancy = opus_rc_p2model(&s->rc, 12);
else if (s->packet.mode == OPUS_MODE_SILK && consumed + 17 <= size * 8)
redundancy = 1;
if (redundancy) {
redundancy_pos = opus_rc_p2model(&s->rc, 1);
if (s->packet.mode == OPUS_MODE_HYBRID)
redundancy_size = opus_rc_unimodel(&s->rc, 256) + 2;
else
redundancy_size = size - (consumed + 7) / 8;
size -= redundancy_size;
if (size < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Invalid redundancy frame size.\n");
return AVERROR_INVALIDDATA;
}
if (redundancy_pos) {
ret = opus_decode_redundancy(s, data + size, redundancy_size);
if (ret < 0)
return ret;
ff_celt_flush(s->celt);
}
}
/* decode the CELT frame */
if (s->packet.mode == OPUS_MODE_CELT || s->packet.mode == OPUS_MODE_HYBRID) {
float *out_tmp[2] = { s->out[0], s->out[1] };
float **dst = (s->packet.mode == OPUS_MODE_CELT) ?
out_tmp : s->celt_output;
int celt_output_samples = samples;
int delay_samples = av_audio_fifo_size(s->celt_delay);
if (delay_samples) {
if (s->packet.mode == OPUS_MODE_HYBRID) {
av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, delay_samples);
for (i = 0; i < s->output_channels; i++) {
s->fdsp->vector_fmac_scalar(out_tmp[i], s->celt_output[i], 1.0,
delay_samples);
out_tmp[i] += delay_samples;
}
celt_output_samples -= delay_samples;
} else {
av_log(s->avctx, AV_LOG_WARNING,
"Spurious CELT delay samples present.\n");
av_audio_fifo_drain(s->celt_delay, delay_samples);
if (s->avctx->err_recognition & AV_EF_EXPLODE)
return AVERROR_BUG;
}
}
opus_raw_init(&s->rc, data + size, size);
ret = ff_celt_decode_frame(s->celt, &s->rc, dst,
s->packet.stereo + 1,
s->packet.frame_duration,
(s->packet.mode == OPUS_MODE_HYBRID) ? 17 : 0,
celt_band_end[s->packet.bandwidth]);
if (ret < 0)
return ret;
if (s->packet.mode == OPUS_MODE_HYBRID) {
int celt_delay = s->packet.frame_duration - celt_output_samples;
void *delaybuf[2] = { s->celt_output[0] + celt_output_samples,
s->celt_output[1] + celt_output_samples
};
for (i = 0; i < s->output_channels; i++) {
s->fdsp->vector_fmac_scalar(out_tmp[i],
s->celt_output[i], 1.0,
celt_output_samples);
}
ret = av_audio_fifo_write(s->celt_delay, delaybuf, celt_delay);
if (ret < 0)
return ret;
}
} else
ff_celt_flush(s->celt);
if (s->redundancy_idx) {
for (i = 0; i < s->output_channels; i++)
opus_fade(s->out[i], s->out[i],
s->redundancy_output[i] + 120 + s->redundancy_idx,
ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
s->redundancy_idx = 0;
}
if (redundancy) {
if (!redundancy_pos) {
ff_celt_flush(s->celt);
ret = opus_decode_redundancy(s, data + size, redundancy_size);
if (ret < 0)
return ret;
for (i = 0; i < s->output_channels; i++) {
opus_fade(s->out[i] + samples - 120 + delayed_samples,
s->out[i] + samples - 120 + delayed_samples,
s->redundancy_output[i] + 120,
ff_celt_window2, 120 - delayed_samples);
if (delayed_samples)
s->redundancy_idx = 120 - delayed_samples;
}
} else {
for (i = 0; i < s->output_channels; i++) {
memcpy(s->out[i] + delayed_samples, s->redundancy_output[i], 120 * sizeof(float));
opus_fade(s->out[i] + 120 + delayed_samples,
s->redundancy_output[i] + 120,
s->out[i] + 120 + delayed_samples,
ff_celt_window2, 120);
}
}
}
return samples;
}
static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size)
{
int samples = s->packet.frame_duration;
int redundancy = 0;
int redundancy_size, redundancy_pos;
int ret, i, consumed;
int delayed_samples = s->delayed_samples;
ret = ff_opus_rc_dec_init(&s->rc, data, size);
if (ret < 0)
return ret;
/* decode the silk frame */
if (s->packet.mode == OPUS_MODE_SILK || s->packet.mode == OPUS_MODE_HYBRID) {
#if CONFIG_SWRESAMPLE
if (!swr_is_initialized(s->swr)) {
#elif CONFIG_AVRESAMPLE
if (!avresample_is_open(s->avr)) {
#endif
ret = opus_init_resample(s);
if (ret < 0)
return ret;
}
samples = ff_silk_decode_superframe(s->silk, &s->rc, s->silk_output,
FFMIN(s->packet.bandwidth, OPUS_BANDWIDTH_WIDEBAND),
s->packet.stereo + 1,
silk_frame_duration_ms[s->packet.config]);
if (samples < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error decoding a SILK frame.\n");
return samples;
}
#if CONFIG_SWRESAMPLE
samples = swr_convert(s->swr,
(uint8_t**)s->out, s->packet.frame_duration,
(const uint8_t**)s->silk_output, samples);
#elif CONFIG_AVRESAMPLE
samples = avresample_convert(s->avr, (uint8_t**)s->out, s->out_size,
s->packet.frame_duration,
(uint8_t**)s->silk_output,
sizeof(s->silk_buf[0]),
samples);
#endif
if (samples < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error resampling SILK data.\n");
return samples;
}
av_assert2((samples & 7) == 0);
s->delayed_samples += s->packet.frame_duration - samples;
} else
ff_silk_flush(s->silk);
// decode redundancy information
consumed = opus_rc_tell(&s->rc);
if (s->packet.mode == OPUS_MODE_HYBRID && consumed + 37 <= size * 8)
redundancy = ff_opus_rc_dec_log(&s->rc, 12);
else if (s->packet.mode == OPUS_MODE_SILK && consumed + 17 <= size * 8)
redundancy = 1;
if (redundancy) {
redundancy_pos = ff_opus_rc_dec_log(&s->rc, 1);
if (s->packet.mode == OPUS_MODE_HYBRID)
redundancy_size = ff_opus_rc_dec_uint(&s->rc, 256) + 2;
else
redundancy_size = size - (consumed + 7) / 8;
size -= redundancy_size;
if (size < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Invalid redundancy frame size.\n");
return AVERROR_INVALIDDATA;
}
if (redundancy_pos) {
ret = opus_decode_redundancy(s, data + size, redundancy_size);
if (ret < 0)
return ret;
ff_celt_flush(s->celt);
}
}
/* decode the CELT frame */
if (s->packet.mode == OPUS_MODE_CELT || s->packet.mode == OPUS_MODE_HYBRID) {
float *out_tmp[2] = { s->out[0], s->out[1] };
float **dst = (s->packet.mode == OPUS_MODE_CELT) ?
out_tmp : s->celt_output;
int celt_output_samples = samples;
int delay_samples = av_audio_fifo_size(s->celt_delay);
if (delay_samples) {
if (s->packet.mode == OPUS_MODE_HYBRID) {
av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, delay_samples);
for (i = 0; i < s->output_channels; i++) {
s->fdsp->vector_fmac_scalar(out_tmp[i], s->celt_output[i], 1.0,
delay_samples);
out_tmp[i] += delay_samples;
}
celt_output_samples -= delay_samples;
} else {
av_log(s->avctx, AV_LOG_WARNING,
"Spurious CELT delay samples present.\n");
av_audio_fifo_drain(s->celt_delay, delay_samples);
if (s->avctx->err_recognition & AV_EF_EXPLODE)
return AVERROR_BUG;
}
}
ff_opus_rc_dec_raw_init(&s->rc, data + size, size);
ret = ff_celt_decode_frame(s->celt, &s->rc, dst,
s->packet.stereo + 1,
s->packet.frame_duration,
(s->packet.mode == OPUS_MODE_HYBRID) ? 17 : 0,
ff_celt_band_end[s->packet.bandwidth]);
if (ret < 0)
return ret;
if (s->packet.mode == OPUS_MODE_HYBRID) {
int celt_delay = s->packet.frame_duration - celt_output_samples;
void *delaybuf[2] = { s->celt_output[0] + celt_output_samples,
s->celt_output[1] + celt_output_samples };
for (i = 0; i < s->output_channels; i++) {
s->fdsp->vector_fmac_scalar(out_tmp[i],
s->celt_output[i], 1.0,
celt_output_samples);
}
ret = av_audio_fifo_write(s->celt_delay, delaybuf, celt_delay);
if (ret < 0)
return ret;
}
} else
ff_celt_flush(s->celt);
if (s->redundancy_idx) {
for (i = 0; i < s->output_channels; i++)
opus_fade(s->out[i], s->out[i],
s->redundancy_output[i] + 120 + s->redundancy_idx,
ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
s->redundancy_idx = 0;
}
if (redundancy) {
if (!redundancy_pos) {
ff_celt_flush(s->celt);
ret = opus_decode_redundancy(s, data + size, redundancy_size);
if (ret < 0)
return ret;
for (i = 0; i < s->output_channels; i++) {
opus_fade(s->out[i] + samples - 120 + delayed_samples,
s->out[i] + samples - 120 + delayed_samples,
s->redundancy_output[i] + 120,
ff_celt_window2, 120 - delayed_samples);
if (delayed_samples)
s->redundancy_idx = 120 - delayed_samples;
}
} else {
for (i = 0; i < s->output_channels; i++) {
memcpy(s->out[i] + delayed_samples, s->redundancy_output[i], 120 * sizeof(float));
opus_fade(s->out[i] + 120 + delayed_samples,
s->redundancy_output[i] + 120,
s->out[i] + 120 + delayed_samples,
ff_celt_window2, 120);
}
}
}
return samples;
}
static int opus_decode_subpacket(OpusStreamContext *s,
const uint8_t *buf, int buf_size,
float **out, int out_size,
int nb_samples)
{
int output_samples = 0;
int flush_needed = 0;
int i, j, ret;
s->out[0] = out[0];
s->out[1] = out[1];
s->out_size = out_size;
/* check if we need to flush the resampler */
#if CONFIG_SWRESAMPLE
if (swr_is_initialized(s->swr)) {
if (buf) {
int64_t cur_samplerate;
av_opt_get_int(s->swr, "in_sample_rate", 0, &cur_samplerate);
flush_needed = (s->packet.mode == OPUS_MODE_CELT) || (cur_samplerate != s->silk_samplerate);
} else {
flush_needed = !!s->delayed_samples;
}
}
#elif CONFIG_AVRESAMPLE
if (avresample_is_open(s->avr)) {
if (buf) {
int64_t cur_samplerate;
av_opt_get_int(s->avr, "in_sample_rate", 0, &cur_samplerate);
flush_needed = (s->packet.mode == OPUS_MODE_CELT) || (cur_samplerate != s->silk_samplerate);
} else {
flush_needed = !!s->delayed_samples;
}
}
#endif
if (!buf && !flush_needed)
return 0;
/* use dummy output buffers if the channel is not mapped to anything */
if (!s->out[0] ||
(s->output_channels == 2 && !s->out[1])) {
av_fast_malloc(&s->out_dummy, &s->out_dummy_allocated_size, s->out_size);
if (!s->out_dummy)
return AVERROR(ENOMEM);
if (!s->out[0])
s->out[0] = s->out_dummy;
if (!s->out[1])
s->out[1] = s->out_dummy;
}
/* flush the resampler if necessary */
if (flush_needed) {
ret = opus_flush_resample(s, s->delayed_samples);
if (ret < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error flushing the resampler.\n");
return ret;
}
#if CONFIG_SWRESAMPLE
swr_close(s->swr);
#elif CONFIG_AVRESAMPLE
avresample_close(s->avr);
#endif
output_samples += s->delayed_samples;
s->delayed_samples = 0;
if (!buf)
goto finish;
}
/* decode all the frames in the packet */
for (i = 0; i < s->packet.frame_count; i++) {
int size = s->packet.frame_size[i];
int samples = opus_decode_frame(s, buf + s->packet.frame_offset[i], size);
if (samples < 0) {
av_log(s->avctx, AV_LOG_ERROR, "Error decoding an Opus frame.\n");
if (s->avctx->err_recognition & AV_EF_EXPLODE)
return samples;
for (j = 0; j < s->output_channels; j++)
memset(s->out[j], 0, s->packet.frame_duration * sizeof(float));
samples = s->packet.frame_duration;
}
output_samples += samples;
for (j = 0; j < s->output_channels; j++)
s->out[j] += samples;
s->out_size -= samples * sizeof(float);
}
finish:
s->out[0] = s->out[1] = NULL;
s->out_size = 0;
return output_samples;
}