diff mbox

[FFmpeg-devel,WIP] libopusdec: Enable FEC/PLC

Message ID E1fIUGT-0003tH-Ql@pannekake.samfundet.no
State New
Headers show

Commit Message

Steinar H. Gunderson May 14, 2018, 10:35 p.m. UTC
Whenever we detect a discontinuity in the incoming stream, ask libopus
to make up intermediate frames based on the first one we actually have
after the discontinuity. If the stream contains FEC data (basically a
low-quality side stream that is delayed by one packet), libopus will
decode that. If not, it will activate PLC (packet loss concealment),
which tries to synthesize some reasonable-sounding frame based on the
previous audio. It will usually be audible, but it's much better than
just playing silence.

Do note that libopus 1.2.1 has a bug that affects PLC for CELT streams,
so you probably want to use Opus from git if you want to test this.

This is a work in progress; in particular, I'm unsure about:

  - Are the samples_to_timebase()/timebase_to_samples() functions
    correct? I've seen avc->pkt_timebase be 0/1 in certain situations,
    which indicates it isn't.
  - Is pts discontinuity the right way of knowing whether packets were
    lost, or can the RTP demuxer signal this somehow? What if the timebase
    conversion is inexact; could we get false positives?
  - Do we need to worry about pkt->pts == AV_NOPTS_VALUE, or can I delete
    the tests in question?

Signed-off-by: Steinar H. Gunderson <steinar+ffmpeg@gunderson.no>
---
 libavcodec/libopusdec.c | 86 +++++++++++++++++++++++++++++++++++------
 1 file changed, 75 insertions(+), 11 deletions(-)

Comments

Michael Niedermayer May 16, 2018, 12:46 a.m. UTC | #1
On Tue, May 15, 2018 at 12:35:45AM +0200, Steinar H. Gunderson wrote:
> Whenever we detect a discontinuity in the incoming stream, ask libopus
> to make up intermediate frames based on the first one we actually have
> after the discontinuity. If the stream contains FEC data (basically a
> low-quality side stream that is delayed by one packet), libopus will
> decode that. If not, it will activate PLC (packet loss concealment),
> which tries to synthesize some reasonable-sounding frame based on the
> previous audio. It will usually be audible, but it's much better than
> just playing silence.
> 

> Do note that libopus 1.2.1 has a bug that affects PLC for CELT streams,
> so you probably want to use Opus from git if you want to test this.

i guess i did hit this

2nd command segfaults:
./ffmpeg -i fate-suite/lossless-audio/luckynight-partial.wma -acodec libopus test.mkv
./ffmpeg -acodec libopus -i test.mkv -f null -
==31996== Invalid write of size 8
==31996==    at 0xCAFA572: opus_multistream_decode_native (opus_multistream_decoder.c:207)
==31996==  Address 0x7fefff6b0 is not stack'd, malloc'd or (recently) free'd
==31996== 
==31996== Invalid write of size 8
==31996==    at 0xCAFA579: opus_multistream_decode_native (opus_multistream_decoder.c:209)
==31996==  Address 0x7ff001438 is not stack'd, malloc'd or (recently) free'd

subsequent tries seem to produce errors instead of segfault

Error while decoding stream #0:0: Unknown error occurred
[libopus @ 0x2a1edc0] Decoding error: buffer too small
Error while decoding stream #0:0: Unknown error occurred


> 
> This is a work in progress; in particular, I'm unsure about:
> 
>   - Are the samples_to_timebase()/timebase_to_samples() functions
>     correct? I've seen avc->pkt_timebase be 0/1 in certain situations,
>     which indicates it isn't.

>   - Is pts discontinuity the right way of knowing whether packets were
>     lost, or can the RTP demuxer signal this somehow? What if the timebase
>     conversion is inexact; could we get false positives?

i would not trust a +-1 difference from expected pts values as indicating a
discontinuity in general. It may be different specifically here, i dont know



>   - Do we need to worry about pkt->pts == AV_NOPTS_VALUE, or can I delete
>     the tests in question?

i would assume pts can be AV_NOPTS_VALUE, 

[...]
Steinar H. Gunderson May 22, 2018, 8:37 p.m. UTC | #2
On Tue, May 15, 2018 at 12:35:45AM +0200, Steinar H. Gunderson wrote:
> Whenever we detect a discontinuity in the incoming stream, ask libopus
> to make up intermediate frames based on the first one we actually have
> after the discontinuity. If the stream contains FEC data (basically a
> low-quality side stream that is delayed by one packet), libopus will
> decode that. If not, it will activate PLC (packet loss concealment),
> which tries to synthesize some reasonable-sounding frame based on the
> previous audio. It will usually be audible, but it's much better than
> just playing silence.

Anyone interested in picking up this?

/* Steinar */
diff mbox

Patch

diff --git a/libavcodec/libopusdec.c b/libavcodec/libopusdec.c
index 2a97811d18..40ee7b8fec 100644
--- a/libavcodec/libopusdec.c
+++ b/libavcodec/libopusdec.c
@@ -43,8 +43,21 @@  struct libopus_context {
 #ifdef OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST
     int apply_phase_inv;
 #endif
+    int64_t expected_next_pts;
 };
 
+const AVRational opus_timebase = { 1, 48000 };
+
+static int samples_to_timebase(const AVCodecContext *avc, int nb_samples)
+{
+    return av_rescale_q(nb_samples, avc->pkt_timebase, opus_timebase);
+}
+
+static int timebase_to_samples(const AVCodecContext *avc, int64_t pts)
+{
+    return av_rescale_q(pts, opus_timebase, avc->pkt_timebase);
+}
+
 #define OPUS_HEAD_SIZE 19
 
 static av_cold int libopus_decode_init(AVCodecContext *avc)
@@ -153,6 +166,8 @@  static av_cold int libopus_decode_init(AVCodecContext *avc)
     /* Decoder delay (in samples) at 48kHz */
     avc->delay = avc->internal->skip_samples = opus->pre_skip;
 
+    opus->expected_next_pts = AV_NOPTS_VALUE;
+
     return 0;
 }
 
@@ -174,25 +189,74 @@  static int libopus_decode(AVCodecContext *avc, void *data,
 {
     struct libopus_context *opus = avc->priv_data;
     AVFrame *frame               = data;
-    int ret, nb_samples;
+    uint8_t *outptr;
+    int ret, nb_samples = 0, nb_lost_samples = 0, nb_samples_left;
+
+    if (opus->expected_next_pts != AV_NOPTS_VALUE &&
+        pkt->pts != AV_NOPTS_VALUE &&
+        pkt->pts != opus->expected_next_pts) {
+        // Cap at recovering 120 ms of lost audio.
+        nb_lost_samples = timebase_to_samples(avc, pkt->pts - opus->expected_next_pts);
+        nb_lost_samples = FFMIN(nb_lost_samples, MAX_FRAME_SIZE);
+    }
 
-    frame->nb_samples = MAX_FRAME_SIZE;
+    frame->nb_samples = MAX_FRAME_SIZE + nb_lost_samples;
     if ((ret = ff_get_buffer(avc, frame, 0)) < 0)
         return ret;
 
+    outptr = frame->data[0];
+    nb_samples_left = frame->nb_samples;
+
+    if (nb_lost_samples) {
+        // Try to recover the lost samples with FEC data from this one.
+        // If there's no FEC data, the decoder will do loss concealment instead.
+        if (avc->sample_fmt == AV_SAMPLE_FMT_S16)
+             nb_samples = opus_multistream_decode(opus->dec, pkt->data, pkt->size,
+                                                  (opus_int16 *)outptr,
+                                                  nb_lost_samples, 1);
+        else
+            nb_samples = opus_multistream_decode_float(opus->dec, pkt->data, pkt->size,
+                                                       (float *)outptr,
+                                                       nb_lost_samples, 1);
+
+        if (nb_samples < 0) {
+            av_log(avc, AV_LOG_ERROR, "Decoding error: %s\n",
+                   opus_strerror(nb_samples));
+            return ff_opus_error_to_averror(nb_samples);
+        }
+
+        av_log(avc, AV_LOG_INFO, "Recovered %d samples\n", nb_samples);
+
+        outptr += nb_samples * avc->channels * av_get_bytes_per_sample(avc->sample_fmt);
+        nb_samples_left -= nb_samples;
+        if (pkt->pts != AV_NOPTS_VALUE) {
+            pkt->pts -= samples_to_timebase(avc, nb_samples);
+            frame->pts = pkt->pts;
+        }
+    }
+
+    // Decode the actual, non-lost data.
     if (avc->sample_fmt == AV_SAMPLE_FMT_S16)
-        nb_samples = opus_multistream_decode(opus->dec, pkt->data, pkt->size,
-                                             (opus_int16 *)frame->data[0],
-                                             frame->nb_samples, 0);
+        ret = opus_multistream_decode(opus->dec, pkt->data, pkt->size,
+                                      (opus_int16 *)outptr,
+                                      nb_samples_left, 0);
     else
-        nb_samples = opus_multistream_decode_float(opus->dec, pkt->data, pkt->size,
-                                                   (float *)frame->data[0],
-                                                   frame->nb_samples, 0);
+        ret = opus_multistream_decode_float(opus->dec, pkt->data, pkt->size,
+                                            (float *)outptr,
+                                            nb_samples_left, 0);
 
-    if (nb_samples < 0) {
+    if (ret < 0) {
         av_log(avc, AV_LOG_ERROR, "Decoding error: %s\n",
-               opus_strerror(nb_samples));
-        return ff_opus_error_to_averror(nb_samples);
+               opus_strerror(ret));
+        return ff_opus_error_to_averror(ret);
+    }
+
+    nb_samples += ret;
+
+    if (pkt->pts == AV_NOPTS_VALUE) {
+        opus->expected_next_pts = AV_NOPTS_VALUE;
+    } else {
+        opus->expected_next_pts = pkt->pts + samples_to_timebase(avc, nb_samples);
     }
 
 #ifndef OPUS_SET_GAIN