gstreamer/ext/opus/gstopusdec.c

905 lines
28 KiB
C

/* GStreamer
* Copyright (C) 2004 Wim Taymans <wim@fluendo.com>
* Copyright (C) 2006 Tim-Philipp Müller <tim centricular net>
* Copyright (C) 2008 Sebastian Dröge <sebastian.droege@collabora.co.uk>
* Copyright (C) 2011-2012 Vincent Penquerc'h <vincent.penquerch@collabora.co.uk>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
/*
* Based on the speexdec element.
*/
/**
* SECTION:element-opusdec
* @see_also: opusenc, oggdemux
*
* This element decodes a OPUS stream to raw integer audio.
*
* <refsect2>
* <title>Example pipelines</title>
* |[
* gst-launch-1.0 -v filesrc location=opus.ogg ! oggdemux ! opusdec ! audioconvert ! audioresample ! alsasink
* ]| Decode an Ogg/Opus file. To create an Ogg/Opus file refer to the documentation of opusenc.
* </refsect2>
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <math.h>
#include <string.h>
#include <stdio.h>
#include "gstopusheader.h"
#include "gstopuscommon.h"
#include "gstopusdec.h"
#include <gst/pbutils/pbutils.h>
GST_DEBUG_CATEGORY_STATIC (opusdec_debug);
#define GST_CAT_DEFAULT opusdec_debug
static GstStaticPadTemplate opus_dec_src_factory =
GST_STATIC_PAD_TEMPLATE ("src",
GST_PAD_SRC,
GST_PAD_ALWAYS,
GST_STATIC_CAPS ("audio/x-raw, "
"format = (string) " GST_AUDIO_NE (S16) ", "
"layout = (string) interleaved, "
"rate = (int) { 48000, 24000, 16000, 12000, 8000 }, "
"channels = (int) [ 1, 8 ] ")
);
static GstStaticPadTemplate opus_dec_sink_factory =
GST_STATIC_PAD_TEMPLATE ("sink",
GST_PAD_SINK,
GST_PAD_ALWAYS,
GST_STATIC_CAPS ("audio/x-opus, "
"channel-mapping-family = (int) 0; "
"audio/x-opus, "
"channel-mapping-family = (int) [1, 255], "
"channels = (int) [1, 255], "
"stream-count = (int) [1, 255], " "coupled-count = (int) [0, 255]")
);
G_DEFINE_TYPE (GstOpusDec, gst_opus_dec, GST_TYPE_AUDIO_DECODER);
#define DB_TO_LINEAR(x) pow (10., (x) / 20.)
#define DEFAULT_USE_INBAND_FEC FALSE
#define DEFAULT_APPLY_GAIN TRUE
enum
{
PROP_0,
PROP_USE_INBAND_FEC,
PROP_APPLY_GAIN
};
static GstFlowReturn gst_opus_dec_parse_header (GstOpusDec * dec,
GstBuffer * buf);
static gboolean gst_opus_dec_start (GstAudioDecoder * dec);
static gboolean gst_opus_dec_stop (GstAudioDecoder * dec);
static GstFlowReturn gst_opus_dec_handle_frame (GstAudioDecoder * dec,
GstBuffer * buffer);
static gboolean gst_opus_dec_set_format (GstAudioDecoder * bdec,
GstCaps * caps);
static void gst_opus_dec_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec);
static void gst_opus_dec_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec);
static GstCaps *gst_opus_dec_getcaps (GstAudioDecoder * dec, GstCaps * filter);
static void
gst_opus_dec_class_init (GstOpusDecClass * klass)
{
GObjectClass *gobject_class;
GstAudioDecoderClass *adclass;
GstElementClass *element_class;
gobject_class = (GObjectClass *) klass;
adclass = (GstAudioDecoderClass *) klass;
element_class = (GstElementClass *) klass;
gobject_class->set_property = gst_opus_dec_set_property;
gobject_class->get_property = gst_opus_dec_get_property;
adclass->start = GST_DEBUG_FUNCPTR (gst_opus_dec_start);
adclass->stop = GST_DEBUG_FUNCPTR (gst_opus_dec_stop);
adclass->handle_frame = GST_DEBUG_FUNCPTR (gst_opus_dec_handle_frame);
adclass->set_format = GST_DEBUG_FUNCPTR (gst_opus_dec_set_format);
adclass->getcaps = GST_DEBUG_FUNCPTR (gst_opus_dec_getcaps);
gst_element_class_add_static_pad_template (element_class,
&opus_dec_src_factory);
gst_element_class_add_static_pad_template (element_class,
&opus_dec_sink_factory);
gst_element_class_set_static_metadata (element_class, "Opus audio decoder",
"Codec/Decoder/Audio", "decode opus streams to audio",
"Vincent Penquerc'h <vincent.penquerch@collabora.co.uk>");
g_object_class_install_property (gobject_class, PROP_USE_INBAND_FEC,
g_param_spec_boolean ("use-inband-fec", "Use in-band FEC",
"Use forward error correction if available (needs PLC enabled)",
DEFAULT_USE_INBAND_FEC, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_APPLY_GAIN,
g_param_spec_boolean ("apply-gain", "Apply gain",
"Apply gain if any is specified in the header", DEFAULT_APPLY_GAIN,
G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS));
GST_DEBUG_CATEGORY_INIT (opusdec_debug, "opusdec", 0,
"opus decoding element");
}
static void
gst_opus_dec_reset (GstOpusDec * dec)
{
dec->packetno = 0;
if (dec->state) {
opus_multistream_decoder_destroy (dec->state);
dec->state = NULL;
}
gst_buffer_replace (&dec->streamheader, NULL);
gst_buffer_replace (&dec->vorbiscomment, NULL);
gst_buffer_replace (&dec->last_buffer, NULL);
dec->primed = FALSE;
dec->pre_skip = 0;
dec->r128_gain = 0;
dec->sample_rate = 0;
dec->n_channels = 0;
dec->leftover_plc_duration = 0;
}
static void
gst_opus_dec_init (GstOpusDec * dec)
{
dec->use_inband_fec = FALSE;
dec->apply_gain = DEFAULT_APPLY_GAIN;
gst_audio_decoder_set_needs_format (GST_AUDIO_DECODER (dec), TRUE);
gst_audio_decoder_set_use_default_pad_acceptcaps (GST_AUDIO_DECODER_CAST
(dec), TRUE);
GST_PAD_SET_ACCEPT_TEMPLATE (GST_AUDIO_DECODER_SINK_PAD (dec));
gst_opus_dec_reset (dec);
}
static gboolean
gst_opus_dec_start (GstAudioDecoder * dec)
{
GstOpusDec *odec = GST_OPUS_DEC (dec);
gst_opus_dec_reset (odec);
/* we know about concealment */
gst_audio_decoder_set_plc_aware (dec, TRUE);
if (odec->use_inband_fec) {
/* opusdec outputs samples directly from an input buffer, except if
* FEC is on, in which case it buffers one buffer in case one buffer
* goes missing.
*/
gst_audio_decoder_set_latency (dec, 120 * GST_MSECOND, 120 * GST_MSECOND);
}
return TRUE;
}
static gboolean
gst_opus_dec_stop (GstAudioDecoder * dec)
{
GstOpusDec *odec = GST_OPUS_DEC (dec);
gst_opus_dec_reset (odec);
return TRUE;
}
static double
gst_opus_dec_get_r128_gain (gint16 r128_gain)
{
return r128_gain / (double) (1 << 8);
}
static double
gst_opus_dec_get_r128_volume (gint16 r128_gain)
{
return DB_TO_LINEAR (gst_opus_dec_get_r128_gain (r128_gain));
}
static gboolean
gst_opus_dec_negotiate (GstOpusDec * dec, const GstAudioChannelPosition * pos)
{
GstCaps *caps = gst_pad_get_allowed_caps (GST_AUDIO_DECODER_SRC_PAD (dec));
GstStructure *s;
GstAudioInfo info;
if (caps) {
gint rate = dec->sample_rate, channels = dec->n_channels;
GstCaps *constraint, *inter;
constraint = gst_caps_from_string ("audio/x-raw");
if (dec->sample_rate == 0 || dec->sample_rate == 48000) {
gst_caps_set_simple (constraint, "rate", G_TYPE_INT, 48000, NULL);
} else {
GValue v = { 0 }
, l = {
0};
g_value_init (&l, GST_TYPE_LIST);
g_value_init (&v, G_TYPE_INT);
g_value_set_int (&v, dec->sample_rate);
gst_value_list_append_value (&l, &v);
g_value_set_int (&v, 48000);
gst_value_list_append_value (&l, &v);
gst_caps_set_value (constraint, "rate", &l);
g_value_unset (&v);
g_value_unset (&l);
}
if (dec->n_channels <= 2) { /* including 0 */
gst_caps_set_simple (constraint, "channels", GST_TYPE_INT_RANGE, 1, 2,
NULL);
} else {
gst_caps_set_simple (constraint, "channels", G_TYPE_INT, dec->n_channels,
NULL);
}
inter = gst_caps_intersect (caps, constraint);
gst_caps_unref (constraint);
if (gst_caps_is_empty (inter)) {
GST_DEBUG_OBJECT (dec, "Empty intersection, failed to negotiate");
gst_caps_unref (inter);
return FALSE;
}
inter = gst_caps_truncate (inter);
s = gst_caps_get_structure (inter, 0);
rate = dec->sample_rate > 0 ? dec->sample_rate : 48000;
gst_structure_fixate_field_nearest_int (s, "rate", dec->sample_rate);
gst_structure_get_int (s, "rate", &rate);
channels = dec->n_channels > 0 ? dec->n_channels : 2;
gst_structure_fixate_field_nearest_int (s, "channels", dec->n_channels);
gst_structure_get_int (s, "channels", &channels);
gst_caps_unref (inter);
dec->sample_rate = rate;
dec->n_channels = channels;
}
if (dec->n_channels == 0) {
GST_DEBUG_OBJECT (dec, "Using a default of 2 channels");
dec->n_channels = 2;
pos = NULL;
}
if (dec->sample_rate == 0) {
GST_DEBUG_OBJECT (dec, "Using a default of 48kHz sample rate");
dec->sample_rate = 48000;
}
GST_INFO_OBJECT (dec, "Negotiated %d channels, %d Hz", dec->n_channels,
dec->sample_rate);
/* pass valid order to audio info */
if (pos) {
memcpy (dec->opus_pos, pos, sizeof (pos[0]) * dec->n_channels);
gst_audio_channel_positions_to_valid_order (dec->opus_pos, dec->n_channels);
}
/* set up source format */
gst_audio_info_init (&info);
gst_audio_info_set_format (&info, GST_AUDIO_FORMAT_S16,
dec->sample_rate, dec->n_channels, pos ? dec->opus_pos : NULL);
gst_audio_decoder_set_output_format (GST_AUDIO_DECODER (dec), &info);
/* but we still need the opus order for later reordering */
if (pos) {
memcpy (dec->opus_pos, pos, sizeof (pos[0]) * dec->n_channels);
} else {
dec->opus_pos[0] = GST_AUDIO_CHANNEL_POSITION_INVALID;
}
dec->info = info;
return TRUE;
}
static GstFlowReturn
gst_opus_dec_parse_header (GstOpusDec * dec, GstBuffer * buf)
{
GstAudioChannelPosition pos[64];
const GstAudioChannelPosition *posn = NULL;
if (!gst_opus_header_is_id_header (buf)) {
GST_ERROR_OBJECT (dec, "Header is not an Opus ID header");
return GST_FLOW_ERROR;
}
if (!gst_codec_utils_opus_parse_header (buf,
&dec->sample_rate,
&dec->n_channels,
&dec->channel_mapping_family,
&dec->n_streams,
&dec->n_stereo_streams,
dec->channel_mapping, &dec->pre_skip, &dec->r128_gain)) {
GST_ERROR_OBJECT (dec, "Failed to parse Opus ID header");
return GST_FLOW_ERROR;
}
dec->r128_gain_volume = gst_opus_dec_get_r128_volume (dec->r128_gain);
GST_INFO_OBJECT (dec,
"Found pre-skip of %u samples, R128 gain %d (volume %f)",
dec->pre_skip, dec->r128_gain, dec->r128_gain_volume);
if (dec->channel_mapping_family == 1) {
GST_INFO_OBJECT (dec, "Channel mapping family 1, Vorbis mapping");
switch (dec->n_channels) {
case 1:
case 2:
/* nothing */
break;
case 3:
case 4:
case 5:
case 6:
case 7:
case 8:
posn = gst_opus_channel_positions[dec->n_channels - 1];
break;
default:{
gint i;
GST_ELEMENT_WARNING (GST_ELEMENT (dec), STREAM, DECODE,
(NULL), ("Using NONE channel layout for more than 8 channels"));
for (i = 0; i < dec->n_channels; i++)
pos[i] = GST_AUDIO_CHANNEL_POSITION_NONE;
posn = pos;
}
}
} else {
GST_INFO_OBJECT (dec, "Channel mapping family %d",
dec->channel_mapping_family);
}
if (!gst_opus_dec_negotiate (dec, posn))
return GST_FLOW_NOT_NEGOTIATED;
return GST_FLOW_OK;
}
static GstFlowReturn
gst_opus_dec_parse_comments (GstOpusDec * dec, GstBuffer * buf)
{
return GST_FLOW_OK;
}
static GstFlowReturn
opus_dec_chain_parse_data (GstOpusDec * dec, GstBuffer * buffer)
{
GstFlowReturn res = GST_FLOW_OK;
gsize size;
guint8 *data;
GstBuffer *outbuf, *bufd;
gint16 *out_data;
int n, err;
int samples;
unsigned int packet_size;
GstBuffer *buf;
GstMapInfo map, omap;
GstAudioClippingMeta *cmeta = NULL;
if (dec->state == NULL) {
/* If we did not get any headers, default to 2 channels */
if (dec->n_channels == 0) {
GST_INFO_OBJECT (dec, "No header, assuming single stream");
dec->n_channels = 2;
dec->sample_rate = 48000;
/* default stereo mapping */
dec->channel_mapping_family = 0;
dec->channel_mapping[0] = 0;
dec->channel_mapping[1] = 1;
dec->n_streams = 1;
dec->n_stereo_streams = 1;
if (!gst_opus_dec_negotiate (dec, NULL))
return GST_FLOW_NOT_NEGOTIATED;
}
if (dec->n_channels == 2 && dec->n_streams == 1
&& dec->n_stereo_streams == 0) {
/* if we are automatically decoding 2 channels, but only have
a single encoded one, direct both channels to it */
dec->channel_mapping[1] = 0;
}
GST_DEBUG_OBJECT (dec, "Creating decoder with %d channels, %d Hz",
dec->n_channels, dec->sample_rate);
#ifndef GST_DISABLE_GST_DEBUG
gst_opus_common_log_channel_mapping_table (GST_ELEMENT (dec), opusdec_debug,
"Mapping table", dec->n_channels, dec->channel_mapping);
#endif
GST_DEBUG_OBJECT (dec, "%d streams, %d stereo", dec->n_streams,
dec->n_stereo_streams);
dec->state =
opus_multistream_decoder_create (dec->sample_rate, dec->n_channels,
dec->n_streams, dec->n_stereo_streams, dec->channel_mapping, &err);
if (!dec->state || err != OPUS_OK)
goto creation_failed;
}
if (buffer) {
GST_DEBUG_OBJECT (dec, "Received buffer of size %" G_GSIZE_FORMAT,
gst_buffer_get_size (buffer));
} else {
GST_DEBUG_OBJECT (dec, "Received missing buffer");
}
/* if using in-band FEC, we introdude one extra frame's delay as we need
to potentially wait for next buffer to decode a missing buffer */
if (dec->use_inband_fec && !dec->primed) {
GST_DEBUG_OBJECT (dec, "First buffer received in FEC mode, early out");
gst_buffer_replace (&dec->last_buffer, buffer);
dec->primed = TRUE;
goto done;
}
/* That's the buffer we'll be sending to the opus decoder. */
buf = (dec->use_inband_fec
&& gst_buffer_get_size (dec->last_buffer) >
0) ? dec->last_buffer : buffer;
/* That's the buffer we get duration from */
bufd = dec->use_inband_fec ? dec->last_buffer : buffer;
if (buf && gst_buffer_get_size (buf) > 0) {
gst_buffer_map (buf, &map, GST_MAP_READ);
data = map.data;
size = map.size;
GST_DEBUG_OBJECT (dec, "Using buffer of size %" G_GSIZE_FORMAT, size);
} else {
/* concealment data, pass NULL as the bits parameters */
GST_DEBUG_OBJECT (dec, "Using NULL buffer");
data = NULL;
size = 0;
}
if (gst_buffer_get_size (bufd) == 0) {
GstClockTime const opus_plc_alignment = 2500 * GST_USECOND;
GstClockTime aligned_missing_duration;
GstClockTime missing_duration = GST_BUFFER_DURATION (bufd);
GST_DEBUG_OBJECT (dec,
"missing buffer, doing PLC duration %" GST_TIME_FORMAT
" plus leftover %" GST_TIME_FORMAT, GST_TIME_ARGS (missing_duration),
GST_TIME_ARGS (dec->leftover_plc_duration));
/* add the leftover PLC duration to that of the buffer */
missing_duration += dec->leftover_plc_duration;
/* align the combined buffer and leftover PLC duration to multiples
* of 2.5ms, rounding to nearest, and store excess duration for later */
aligned_missing_duration =
((missing_duration +
opus_plc_alignment / 2) / opus_plc_alignment) * opus_plc_alignment;
dec->leftover_plc_duration = missing_duration - aligned_missing_duration;
/* Opus' PLC cannot operate with less than 2.5ms; skip PLC
* and accumulate the missing duration in the leftover_plc_duration
* for the next PLC attempt */
if (aligned_missing_duration < opus_plc_alignment) {
GST_DEBUG_OBJECT (dec,
"current duration %" GST_TIME_FORMAT
" of missing data not enough for PLC (minimum needed: %"
GST_TIME_FORMAT ") - skipping", GST_TIME_ARGS (missing_duration),
GST_TIME_ARGS (opus_plc_alignment));
goto done;
}
/* convert the duration (in nanoseconds) to sample count */
samples =
gst_util_uint64_scale_int (aligned_missing_duration, dec->sample_rate,
GST_SECOND);
GST_DEBUG_OBJECT (dec,
"calculated PLC frame length: %" GST_TIME_FORMAT
" num frame samples: %d new leftover: %" GST_TIME_FORMAT,
GST_TIME_ARGS (aligned_missing_duration), samples,
GST_TIME_ARGS (dec->leftover_plc_duration));
} else {
/* use maximum size (120 ms) as the number of returned samples is
not constant over the stream. */
samples = 120 * dec->sample_rate / 1000;
}
packet_size = samples * dec->n_channels * 2;
outbuf =
gst_audio_decoder_allocate_output_buffer (GST_AUDIO_DECODER (dec),
packet_size);
if (!outbuf) {
goto buffer_failed;
}
gst_buffer_map (outbuf, &omap, GST_MAP_WRITE);
out_data = (gint16 *) omap.data;
if (dec->use_inband_fec) {
if (gst_buffer_get_size (dec->last_buffer) > 0) {
/* normal delayed decode */
GST_LOG_OBJECT (dec, "FEC enabled, decoding last delayed buffer");
n = opus_multistream_decode (dec->state, data, size, out_data, samples,
0);
} else {
/* FEC reconstruction decode */
GST_LOG_OBJECT (dec, "FEC enabled, reconstructing last buffer");
n = opus_multistream_decode (dec->state, data, size, out_data, samples,
1);
}
} else {
/* normal decode */
GST_LOG_OBJECT (dec, "FEC disabled, decoding buffer");
n = opus_multistream_decode (dec->state, data, size, out_data, samples, 0);
}
gst_buffer_unmap (outbuf, &omap);
if (data != NULL)
gst_buffer_unmap (buf, &map);
if (n < 0) {
GST_ELEMENT_ERROR (dec, STREAM, DECODE, ("Decoding error: %d", n), (NULL));
gst_buffer_unref (outbuf);
return GST_FLOW_ERROR;
}
GST_DEBUG_OBJECT (dec, "decoded %d samples", n);
gst_buffer_set_size (outbuf, n * 2 * dec->n_channels);
cmeta = gst_buffer_get_audio_clipping_meta (buf);
g_assert (!cmeta || cmeta->format == GST_FORMAT_DEFAULT);
/* Skip any samples that need skipping */
if (cmeta && cmeta->start) {
guint pre_skip = cmeta->start;
guint scaled_pre_skip = pre_skip * dec->sample_rate / 48000;
guint skip = scaled_pre_skip > n ? n : scaled_pre_skip;
guint scaled_skip = skip * 48000 / dec->sample_rate;
gst_buffer_resize (outbuf, skip * 2 * dec->n_channels, -1);
GST_INFO_OBJECT (dec,
"Skipping %u samples at the beginning (%u at 48000 Hz)",
skip, scaled_skip);
}
if (cmeta && cmeta->end) {
guint post_skip = cmeta->end;
guint scaled_post_skip = post_skip * dec->sample_rate / 48000;
guint skip = scaled_post_skip > n ? n : scaled_post_skip;
guint scaled_skip = skip * 48000 / dec->sample_rate;
guint outsize = gst_buffer_get_size (outbuf);
guint skip_bytes = skip * 2 * dec->n_channels;
if (outsize > skip_bytes)
outsize -= skip_bytes;
else
outsize = 0;
gst_buffer_resize (outbuf, 0, outsize);
GST_INFO_OBJECT (dec,
"Skipping %u samples at the end (%u at 48000 Hz)", skip, scaled_skip);
}
if (gst_buffer_get_size (outbuf) == 0) {
gst_buffer_unref (outbuf);
outbuf = NULL;
} else if (dec->opus_pos[0] != GST_AUDIO_CHANNEL_POSITION_INVALID) {
gst_audio_buffer_reorder_channels (outbuf, GST_AUDIO_FORMAT_S16,
dec->n_channels, dec->opus_pos, dec->info.position);
}
/* Apply gain */
/* Would be better off leaving this to a volume element, as this is
a naive conversion that does too many int/float conversions.
However, we don't have control over the pipeline...
So make it optional if the user program wants to use a volume,
but do it by default so the correct volume goes out by default */
if (dec->apply_gain && outbuf && dec->r128_gain) {
gsize rsize;
unsigned int i, nsamples;
double volume = dec->r128_gain_volume;
gint16 *samples;
gst_buffer_map (outbuf, &omap, GST_MAP_READWRITE);
samples = (gint16 *) omap.data;
rsize = omap.size;
GST_DEBUG_OBJECT (dec, "Applying gain: volume %f", volume);
nsamples = rsize / 2;
for (i = 0; i < nsamples; ++i) {
int sample = (int) (samples[i] * volume + 0.5);
samples[i] = sample < -32768 ? -32768 : sample > 32767 ? 32767 : sample;
}
gst_buffer_unmap (outbuf, &omap);
}
if (dec->use_inband_fec) {
gst_buffer_replace (&dec->last_buffer, buffer);
}
res = gst_audio_decoder_finish_frame (GST_AUDIO_DECODER (dec), outbuf, 1);
if (res != GST_FLOW_OK)
GST_DEBUG_OBJECT (dec, "flow: %s", gst_flow_get_name (res));
done:
return res;
creation_failed:
GST_ERROR_OBJECT (dec, "Failed to create Opus decoder: %d", err);
return GST_FLOW_ERROR;
buffer_failed:
GST_ERROR_OBJECT (dec, "Failed to create %u byte buffer", packet_size);
return GST_FLOW_ERROR;
}
static gboolean
gst_opus_dec_set_format (GstAudioDecoder * bdec, GstCaps * caps)
{
GstOpusDec *dec = GST_OPUS_DEC (bdec);
gboolean ret = TRUE;
GstStructure *s;
const GValue *streamheader;
GstCaps *old_caps;
GST_DEBUG_OBJECT (dec, "set_format: %" GST_PTR_FORMAT, caps);
if ((old_caps = gst_pad_get_current_caps (GST_AUDIO_DECODER_SINK_PAD (bdec)))) {
if (gst_caps_is_equal (caps, old_caps)) {
gst_caps_unref (old_caps);
GST_DEBUG_OBJECT (dec, "caps didn't change");
goto done;
}
GST_DEBUG_OBJECT (dec, "caps have changed, resetting decoder");
gst_opus_dec_reset (dec);
gst_caps_unref (old_caps);
}
s = gst_caps_get_structure (caps, 0);
if ((streamheader = gst_structure_get_value (s, "streamheader")) &&
G_VALUE_HOLDS (streamheader, GST_TYPE_ARRAY) &&
gst_value_array_get_size (streamheader) >= 2) {
const GValue *header, *vorbiscomment;
GstBuffer *buf;
GstFlowReturn res = GST_FLOW_OK;
header = gst_value_array_get_value (streamheader, 0);
if (header && G_VALUE_HOLDS (header, GST_TYPE_BUFFER)) {
buf = gst_value_get_buffer (header);
res = gst_opus_dec_parse_header (dec, buf);
if (res != GST_FLOW_OK) {
ret = FALSE;
goto done;
}
gst_buffer_replace (&dec->streamheader, buf);
}
vorbiscomment = gst_value_array_get_value (streamheader, 1);
if (vorbiscomment && G_VALUE_HOLDS (vorbiscomment, GST_TYPE_BUFFER)) {
buf = gst_value_get_buffer (vorbiscomment);
res = gst_opus_dec_parse_comments (dec, buf);
if (res != GST_FLOW_OK) {
ret = FALSE;
goto done;
}
gst_buffer_replace (&dec->vorbiscomment, buf);
}
} else {
const GstAudioChannelPosition *posn = NULL;
if (!gst_codec_utils_opus_parse_caps (caps, &dec->sample_rate,
&dec->n_channels, &dec->channel_mapping_family, &dec->n_streams,
&dec->n_stereo_streams, dec->channel_mapping)) {
ret = FALSE;
goto done;
}
if (dec->channel_mapping_family == 1 && dec->n_channels <= 8)
posn = gst_opus_channel_positions[dec->n_channels - 1];
if (!gst_opus_dec_negotiate (dec, posn))
return FALSE;
}
done:
return ret;
}
static gboolean
memcmp_buffers (GstBuffer * buf1, GstBuffer * buf2)
{
gsize size1, size2;
gboolean res;
GstMapInfo map;
size1 = gst_buffer_get_size (buf1);
size2 = gst_buffer_get_size (buf2);
if (size1 != size2)
return FALSE;
gst_buffer_map (buf1, &map, GST_MAP_READ);
res = gst_buffer_memcmp (buf2, 0, map.data, map.size) == 0;
gst_buffer_unmap (buf1, &map);
return res;
}
static GstFlowReturn
gst_opus_dec_handle_frame (GstAudioDecoder * adec, GstBuffer * buf)
{
GstFlowReturn res;
GstOpusDec *dec;
/* no fancy draining */
if (G_UNLIKELY (!buf))
return GST_FLOW_OK;
dec = GST_OPUS_DEC (adec);
GST_LOG_OBJECT (dec,
"Got buffer ts %" GST_TIME_FORMAT ", duration %" GST_TIME_FORMAT,
GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (buf)),
GST_TIME_ARGS (GST_BUFFER_DURATION (buf)));
/* If we have the streamheader and vorbiscomment from the caps already
* ignore them here */
if (dec->streamheader && dec->vorbiscomment) {
if (memcmp_buffers (dec->streamheader, buf)) {
GST_DEBUG_OBJECT (dec, "found streamheader");
gst_audio_decoder_finish_frame (adec, NULL, 1);
res = GST_FLOW_OK;
} else if (memcmp_buffers (dec->vorbiscomment, buf)) {
GST_DEBUG_OBJECT (dec, "found vorbiscomments");
gst_audio_decoder_finish_frame (adec, NULL, 1);
res = GST_FLOW_OK;
} else {
res = opus_dec_chain_parse_data (dec, buf);
}
} else {
/* Otherwise fall back to packet counting and assume that the
* first two packets might be the headers, checking magic. */
switch (dec->packetno) {
case 0:
if (gst_opus_header_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);
}
break;
case 1:
if (gst_opus_header_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);
}
break;
default:
{
res = opus_dec_chain_parse_data (dec, buf);
break;
}
}
}
dec->packetno++;
return res;
}
static void
gst_opus_dec_get_property (GObject * object, guint prop_id, GValue * value,
GParamSpec * pspec)
{
GstOpusDec *dec = GST_OPUS_DEC (object);
switch (prop_id) {
case PROP_USE_INBAND_FEC:
g_value_set_boolean (value, dec->use_inband_fec);
break;
case PROP_APPLY_GAIN:
g_value_set_boolean (value, dec->apply_gain);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_opus_dec_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec)
{
GstOpusDec *dec = GST_OPUS_DEC (object);
switch (prop_id) {
case PROP_USE_INBAND_FEC:
dec->use_inband_fec = g_value_get_boolean (value);
break;
case PROP_APPLY_GAIN:
dec->apply_gain = g_value_get_boolean (value);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
GstCaps *
gst_opus_dec_getcaps (GstAudioDecoder * dec, GstCaps * filter)
{
GstCaps *caps;
unsigned n;
int channels;
if (filter) {
filter = gst_caps_copy (filter);
for (n = 0; n < gst_caps_get_size (filter); ++n) {
GstStructure *s = gst_caps_get_structure (filter, n);
if (gst_structure_get_int (s, "channels", &channels)) {
if (channels == 1 || channels == 2) {
GValue v = { 0 };
g_value_init (&v, GST_TYPE_INT_RANGE);
gst_value_set_int_range (&v, 1, 2);
gst_structure_set_value (s, "channels", &v);
g_value_unset (&v);
}
}
}
}
caps = gst_audio_decoder_proxy_getcaps (dec, NULL, filter);
if (filter)
gst_caps_unref (filter);
if (caps) {
caps = gst_caps_make_writable (caps);
for (n = 0; n < gst_caps_get_size (caps); ++n) {
GstStructure *s = gst_caps_get_structure (caps, n);
if (gst_structure_get_int (s, "channels", &channels)) {
if (channels == 1 || channels == 2) {
GValue v = { 0 };
g_value_init (&v, GST_TYPE_INT_RANGE);
gst_value_set_int_range (&v, 1, 2);
gst_structure_set_value (s, "channels", &v);
g_value_unset (&v);
}
}
}
}
return caps;
}