opusdec: rewrite logic

Parameters such as frame size, etc, are variable. Pretty much
everything can change within a stream, so be prepared about it,
and do not cache parameters in the decoder.
This commit is contained in:
Vincent Penquerc'h 2011-11-16 01:14:32 +00:00
parent 9e79a8ed01
commit 70ca2a6851
2 changed files with 53 additions and 87 deletions

View file

@ -48,9 +48,6 @@
GST_DEBUG_CATEGORY_STATIC (opusdec_debug);
#define GST_CAT_DEFAULT opusdec_debug
#define DEC_MAX_FRAME_SIZE 2000
#define DEC_MAX_OUTPUT_BUFFER_SIZE (5760)
static GstStaticPadTemplate opus_dec_src_factory =
GST_STATIC_PAD_TEMPLATE ("src",
GST_PAD_SRC,
@ -110,14 +107,13 @@ static void
gst_opus_dec_reset (GstOpusDec * dec)
{
dec->packetno = 0;
dec->frame_size = 0;
dec->frame_samples = 960;
dec->frame_duration = 0;
if (dec->state) {
opus_decoder_destroy (dec->state);
dec->state = NULL;
}
dec->next_ts = 0;
gst_buffer_replace (&dec->streamheader, NULL);
gst_buffer_replace (&dec->vorbiscomment, NULL);
}
@ -176,7 +172,7 @@ opus_dec_chain_parse_data (GstOpusDec * dec, GstBuffer * buf,
GstBuffer *outbuf;
gint16 *out_data;
int n, err;
int samples_per_frame;
int samples;
unsigned int packet_size;
if (dec->state == NULL) {
@ -192,8 +188,8 @@ opus_dec_chain_parse_data (GstOpusDec * dec, GstBuffer * buf,
"rate", G_TYPE_INT, dec->sample_rate,
"channels", G_TYPE_INT, dec->n_channels, NULL);
GST_DEBUG_OBJECT (dec, "rate=%d channels=%d frame-size=%d",
dec->sample_rate, dec->n_channels, dec->frame_size);
GST_DEBUG_OBJECT (dec, "rate=%d channels=%d",
dec->sample_rate, dec->n_channels);
if (!gst_pad_set_caps (GST_AUDIO_DECODER_SRC_PAD (dec), caps))
GST_ERROR ("nego failure");
@ -214,12 +210,13 @@ opus_dec_chain_parse_data (GstOpusDec * dec, GstBuffer * buf,
size = 0;
}
samples_per_frame =
opus_packet_get_samples_per_frame (data, dec->sample_rate);
samples =
opus_packet_get_samples_per_frame (data,
dec->sample_rate) * opus_packet_get_nb_frames (data, size);
GST_DEBUG ("bandwidth %d", opus_packet_get_bandwidth (data));
GST_DEBUG ("samples_per_frame %d", samples_per_frame);
GST_DEBUG ("samples %d", samples);
packet_size = samples_per_frame * dec->n_channels * 2;
packet_size = samples * dec->n_channels * 2;
outbuf = gst_buffer_new_and_alloc (packet_size);
if (!outbuf) {
goto buffer_failed;
@ -227,34 +224,33 @@ opus_dec_chain_parse_data (GstOpusDec * dec, GstBuffer * buf,
out_data = (gint16 *) gst_buffer_map (outbuf, &out_size, NULL, GST_MAP_WRITE);
GST_LOG_OBJECT (dec, "decoding frame");
GST_LOG_OBJECT (dec, "decoding %d samples, in size %u", samples, size);
n = opus_decode (dec->state, data, size, out_data, dec->frame_samples, 0);
n = opus_decode (dec->state, data, size, out_data, samples, 0);
gst_buffer_unmap (buf, data, size);
gst_buffer_unmap (outbuf, out_data, out_size);
if (n < 0) {
gst_buffer_unmap (outbuf, out_data, out_size);
GST_ELEMENT_ERROR (dec, STREAM, DECODE, ("Decoding error: %d", n), (NULL));
return GST_FLOW_ERROR;
}
GST_DEBUG_OBJECT (dec, "decoded %d samples", n);
if (!GST_CLOCK_TIME_IS_VALID (timestamp)) {
GST_WARNING_OBJECT (dec, "No timestamp in -> no timestamp out");
if (GST_CLOCK_TIME_IS_VALID (timestamp)) {
GST_BUFFER_TIMESTAMP (outbuf) = timestamp;
} else {
GST_BUFFER_TIMESTAMP (outbuf) = dec->next_ts;
}
GST_DEBUG_OBJECT (dec, "timestamp=%" GST_TIME_FORMAT,
GST_TIME_ARGS (timestamp));
GST_BUFFER_TIMESTAMP (outbuf) = GST_BUFFER_TIMESTAMP (buf);
GST_BUFFER_DURATION (outbuf) = GST_BUFFER_DURATION (buf);
GST_BUFFER_DURATION (outbuf) =
gst_util_uint64_scale (n, GST_SECOND, dec->sample_rate);
dec->next_ts = GST_BUFFER_TIMESTAMP (outbuf) + GST_BUFFER_DURATION (outbuf);
GST_LOG_OBJECT (dec, "pushing buffer with ts=%" GST_TIME_FORMAT ", dur=%"
GST_TIME_FORMAT, GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (outbuf)),
GST_TIME_ARGS (dec->frame_duration));
GST_TIME_ARGS (GST_BUFFER_DURATION (outbuf)));
res = gst_audio_decoder_finish_frame (GST_AUDIO_DECODER (dec), outbuf, 1);
gst_buffer_unmap (outbuf, out_data, out_size);
if (res != GST_FLOW_OK)
GST_DEBUG_OBJECT (dec, "flow: %s", gst_flow_get_name (res));
@ -269,37 +265,6 @@ buffer_failed:
return GST_FLOW_ERROR;
}
static gint
gst_opus_dec_get_frame_samples (GstOpusDec * dec)
{
gint frame_samples = 0;
switch (dec->frame_size) {
case 2:
frame_samples = dec->sample_rate / 400;
break;
case 5:
frame_samples = dec->sample_rate / 200;
break;
case 10:
frame_samples = dec->sample_rate / 100;
break;
case 20:
frame_samples = dec->sample_rate / 50;
break;
case 40:
frame_samples = dec->sample_rate / 25;
break;
case 60:
frame_samples = 3 * dec->sample_rate / 50;
break;
default:
GST_WARNING_OBJECT (dec, "Unsupported frame size: %d", dec->frame_size);
frame_samples = 0;
break;
}
return frame_samples;
}
static gboolean
gst_opus_dec_set_format (GstAudioDecoder * bdec, GstCaps * caps)
{
@ -337,25 +302,6 @@ gst_opus_dec_set_format (GstAudioDecoder * bdec, GstCaps * caps)
}
}
if (!gst_structure_get_int (s, "frame-size", &dec->frame_size)) {
GST_WARNING_OBJECT (dec, "Frame size not included in caps");
}
if (!gst_structure_get_int (s, "channels", &dec->n_channels)) {
GST_WARNING_OBJECT (dec, "Number of channels not included in caps");
}
if (!gst_structure_get_int (s, "rate", &dec->sample_rate)) {
GST_WARNING_OBJECT (dec, "Sample rate not included in caps");
}
dec->frame_samples = gst_opus_dec_get_frame_samples (dec);
dec->frame_duration = gst_util_uint64_scale_int (dec->frame_samples,
GST_SECOND, dec->sample_rate);
GST_INFO_OBJECT (dec,
"Got frame size %d, %d channels, %d Hz, giving %d samples per frame, frame duration %"
GST_TIME_FORMAT, dec->frame_size, dec->n_channels, dec->sample_rate,
dec->frame_samples, GST_TIME_ARGS (dec->frame_duration));
caps = gst_caps_new_simple ("audio/x-raw",
"format", G_TYPE_STRING, "S16LE",
"rate", G_TYPE_INT, dec->sample_rate,
@ -387,6 +333,20 @@ memcmp_buffers (GstBuffer * buf1, GstBuffer * buf2)
return res;
}
static gboolean
gst_opus_dec_is_header (GstBuffer * buf, const char *magic, guint magic_size)
{
guint8 *data;
gsize size;
gboolean ret;
data = gst_buffer_map (buf, &size, NULL, GST_MAP_READ);
if (!data)
return FALSE;
ret = (size >= magic_size && !memcmp (magic, data, magic_size));
gst_buffer_unmap (buf, data, size);
return ret;
}
static GstFlowReturn
gst_opus_dec_handle_frame (GstAudioDecoder * adec, GstBuffer * buf)
@ -424,16 +384,24 @@ gst_opus_dec_handle_frame (GstAudioDecoder * adec, GstBuffer * buf)
* first two packets are the headers. */
switch (dec->packetno) {
case 0:
GST_DEBUG_OBJECT (dec, "counted streamheader");
res = GST_FLOW_OK;
res = gst_opus_dec_parse_header (dec, buf);
gst_audio_decoder_finish_frame (adec, NULL, 1);
if (gst_opus_dec_is_header (buf, "OpusHead", 8)) {
GST_DEBUG_OBJECT (dec, "found streamheader");
res = gst_opus_dec_parse_header (dec, buf);
gst_audio_decoder_finish_frame (adec, NULL, 1);
} else {
res = opus_dec_chain_parse_data (dec, buf, GST_BUFFER_TIMESTAMP (buf),
GST_BUFFER_DURATION (buf));
}
break;
case 1:
GST_DEBUG_OBJECT (dec, "counted vorbiscomments");
res = GST_FLOW_OK;
res = gst_opus_dec_parse_comments (dec, buf);
gst_audio_decoder_finish_frame (adec, NULL, 1);
if (gst_opus_dec_is_header (buf, "OpusTags", 8)) {
GST_DEBUG_OBJECT (dec, "counted vorbiscomments");
res = gst_opus_dec_parse_comments (dec, buf);
gst_audio_decoder_finish_frame (adec, NULL, 1);
} else {
res = opus_dec_chain_parse_data (dec, buf, GST_BUFFER_TIMESTAMP (buf),
GST_BUFFER_DURATION (buf));
}
break;
default:
{

View file

@ -45,11 +45,9 @@ struct _GstOpusDec {
GstAudioDecoder element;
OpusDecoder *state;
int frame_samples;
gint frame_size;
GstClockTime frame_duration;
guint64 packetno;
GstClockTime next_ts;
GstBuffer *streamheader;
GstBuffer *vorbiscomment;