Cleanup avformat-based preprocessors #2665

lu-zero · 2021-05-15T13:20:34Z

If you like the set I can convert the other modules likewise and further factor out the common parts.

januscla · 2021-05-15T13:20:36Z

Thanks for your contribution, @lu-zero! Please make sure you sign our CLA, as it's a required step before we can merge this.

lminiero · 2021-05-17T06:59:16Z

Thanks for the contribution @lu-zero! I definitely see value in the more compact approach you're suggesting, since we had indeed many similar sections that could be streamlined.

I only have one doubt related to the changes you made on AVPacket management, probably related to my limited knowledge of its memory management internals: I see the packet is allocated once before the loop is started, and then at each loop there's a av_packet_unref call that precedes its usage. Won't that first unref actually get rid of the packet before we get a chance of using it? First iteration in the loop aside, it's also not clear whether av_write_frame will add a reference to the packet when using it, as the documentation says it does not take ownership.

There is no need to allocate an encoder to mux. Makes possible to use the postprocessor without having a vp8/vp9 encoder in avcodec.

lu-zero · 2021-05-17T08:50:18Z

packet_unref() does the same work as the former init_packet(). Since we do not use av_buffer, there is no actual refcounting on the payload.

void av_packet_unref(AVPacket *pkt)
{
    av_packet_free_side_data(pkt); // NO-OP we do not have side data
    av_buffer_unref(&pkt->buf); // NO-OP we do not use the buffer
    av_init_packet(pkt); // same as the deprecated function
    pkt->data = NULL; // basic sanitization
    pkt->size = 0;
}

lminiero · 2021-05-17T15:32:01Z

Ack, thanks for the clarification 👍

lu-zero · 2021-05-19T13:50:18Z

@lminiero do you have test samples I can use to make sure everything works after the changes? I have samples for opus and vp8 but not for the others.

lminiero · 2021-05-19T15:29:23Z

@lu-zero I don't have any recording in the other formats, but they should be quite easy to generate. If you have a Janus instance you control somewhere, you can open the echotest demo with a custom codec preference, e.g.:

https://.../echotest.html?vcodec=h264

to force a specific video codec to be used (acodec does the same for audio), and then on the JS console send a record: true command to start recording the session:

echotest.send({ message: { record: true }});

This will create a different mjr file for each medium (in EchoTest audio, video and data). Sending record: false or stopping the demo will finalize the files, and allow you to use them.

lminiero · 2021-05-21T16:35:58Z

@lu-zero please let me know when you think the refactoring is done, so that I know when I can do a review 👍
I'm currently working on a separate branch that, due to some new features, will touch the way some of the packets are processed in the pprec tool as well, so I'll also need to know what I'll have to take into account in terms of refactoring there.

lu-zero · 2021-05-21T16:51:17Z

Do you have a known use for the silence packets in opus? I just commented it out since locally it seems to cause more problems that it solves. If you are happy with what's in there I'd say it could land.

lminiero · 2021-05-21T17:08:35Z

Do you have a known use for the silence packets in opus? I just commented it out since locally it seems to cause more problems that it solves. If you are happy with what's in there I'd say it could land.

Actually yes, that's quite important. When we detect a packet loss, we have to inject a silence packet, otherwise the duration of the processed stream will be broken. That's because, at least from my understanding of libogg, packets can only be added to the format with monotonically increasing packet IDs, and so skipping missing packets reduces the duration of the stream (as the 20ms that were missed are not reflected in the audio recording). That might not be important in case of very few losses, but if there's many of them, this could lead to problems, like out-of-sync audio and video streams if they need to be combined ex-post.

lminiero

Just had a look at the code, and it looks like a really useful refactoring, thanks! I just added a few minor editorial comments inline (mostly related to the project code style), and a note on the Opus silence ingestion we discussed a few days ago. I haven't tested this yet, but plan to do it soon.

lminiero · 2021-05-24T09:17:50Z

Makefile.am

@@ -585,6 +585,8 @@ janus_pp_rec_SOURCES = \
 	postprocessing/pp-h264.h \
 	postprocessing/pp-av1.c \
 	postprocessing/pp-av1.h \
+	postprocessing/pp-avformat.h \
+	postprocessing/pp-avformat.c \


Just a small nit, can you invert the order of .c and .h file here to match the others in the list?

lminiero · 2021-05-24T09:19:30Z

postprocessing/pp-av1.c

-#else
-	vStream = avformat_new_stream(fctx, 0);
+
+	vStream = janus_pp_new_video_avstream(fctx, AV_CODEC_ID_AV1, max_width, max_height);;


Small nit: there's a double semicolon at the end.

lminiero · 2021-05-24T09:22:27Z

postprocessing/pp-avformat.c

+
+	/* We save the metadata part as a comment (see #1189) */
+        if(metadata)
+                av_dict_set(&ctx->metadata, "comment", metadata, 0);


Broken indentation in the above lines: per the project code style, we use tabs, not spaces.

lminiero · 2021-05-24T09:22:45Z

postprocessing/pp-avformat.c

+                JANUS_LOG(LOG_ERR, "Error guessing format\n");
+		avformat_free_context(ctx);
+                return NULL;
+        }


Same indentation issues for the above lines.

lminiero · 2021-05-24T09:27:57Z

postprocessing/pp-opus.c


+	/* WebM output */
+	fctx = janus_pp_create_avformatcontext("ogg", metadata, destination);


I guess that if we wanted to optionally allow people to output the processing to .mka, as you said in #2658, we'd only need to make the "ogg" part a configurable string where we can put a different format target, correct? Same for videos, I guess, where VP8/VP9 could be outputted to .mkv rather than .webm as we do now.

That's the plan, ideally should be easy to select the output format (mp4 or mkv or webm).

I'll try to make the postprocessor able to mux both audio and video in a single pass later probably. One step at time first though :)

I'll try to make the postprocessor able to mux both audio and video in a single pass later probably. One step at time first though :)

That's actually not needed, since our .mjr files only store one media stream anyway (audio, video or data). Any muxing after that should be done separately, e.g., using ffmpeg itself, and possibly using timing info from the .mjr header to sort out sync issues.

lminiero · 2021-05-24T09:29:10Z

postprocessing/pp-opus.c

+		av_packet_unref(pkt);
+		pkt->stream_index = 0;
+		pkt->data = buffer;
+                pkt->size = bytes;


Broken indentation (spaces).

Looks like in one of the machines I tested I do not have editorconfig support. Do you happen to have a clang-format setup I can use?

lminiero · 2021-05-24T09:30:52Z

postprocessing/pp-opus.c

+		pkt->stream_index = 0;
+		pkt->data = buffer;
+                pkt->size = bytes;
+		pkt->pts = pkt->dts = av_rescale_q(tmp->ts - list->ts, timebase, fctx->streams[0]->time_base);


Is this maybe why you were getting issues with the silence packets being inserted manually as we did when using libogg directly? Does the ogg format target in ffmpeg automatically insert silence audio packets looking at what the pts is, or is still increased monotonically even in case of gaps?

The sample I used to test has the audio lasting much longer than the video, I thought the problem was in the audio, but probably it is the video having problems.

I still need to get a local setup to get better samples.

Inserting the silence seems fine with both ogg and mkv output.

Inserting the silence seems fine with both ogg and mkv output.

Ack, thanks!

I still need to get a local setup to get better samples.

There are a couple of old .mjr files you can play with in this folder of the repo: they use the old format, but should be ok if you want to test audio and video streams that have the same length.

lminiero · 2021-05-24T09:32:23Z

postprocessing/pp-webm.c

 	#else
-	vStream->codec->codec_id = AV_CODEC_ID_VP8;
+		codec_id = AV_CODEC_ID_VP8;


The two codec_id lines don't need to be indented (the one below isn't, for instance), as macro ifs for us are not like "regular" ifs.

lu-zero · 2021-05-24T11:56:45Z

The formatting problems should be addressed now, I hope :)

lminiero · 2021-05-24T13:28:37Z

postprocessing/pp-opus.c

+    'O', 'p', 'u', 's', 'H', 'e', 'a', 'd',
+    1, 2, 56, 1, 128, 187,
+    0, 0, 0, 0, 0,
+};


Whoops looks like I missed this space-indentation in the previous review, sorry...

lminiero · 2021-05-24T13:31:32Z

The formatting problems should be addressed now, I hope :)

Thanks! I just found one I missed before (I added a comment above), but the code changes do look fine to me. I plan to make some tests either later today or tomorrow morning, and in case they work fine for me it will be good to merge. Please make sure you sign our CLA before we get to that, though, as we won't be able to merge otherwise.

lu-zero · 2021-05-24T13:32:37Z

I think I already signed the CLA, it does not show on your side?

lminiero · 2021-05-24T13:34:08Z

Mh no, it's not there. Did you fill the form at the end of the page? Maybe you only did the Github authentication on top? (the page is indeed a bit confusing in that regard)

lu-zero · 2021-05-24T13:59:01Z

I completed the form at least now.

lminiero · 2021-05-24T14:01:22Z

Thanks, I do see it now! ✌️

lminiero · 2021-05-25T09:57:31Z

postprocessing/pp-g722.c

-		av_init_packet(&avpacket);
-		avpacket.data = (uint8_t *)buffer;
-		avpacket.size = bytes;
+		AVPacket *avpacket = av_packet_alloc();


@lu-zero any reason why this allocation is inside the loop? I see for all other codecs you just do it once before the while, and just unref/initialize the packet at each loop, while for G.722 this is happening in the while instead. Is it because we're actually decoding something here, rather than just saving to a container?

lu-zero · 2021-05-25T10:21:10Z

Yes, it felt simpler that way.

…

On Tue, May 25, 2021, 11:57 Lorenzo Miniero ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In postprocessing/pp-g722.c <#2665 (comment)> : > @@ -192,18 +170,17 @@ int janus_pp_g722_process(FILE *file, janus_pp_frame_packet *list, int *working) JANUS_LOG(LOG_VERB, "Writing %d bytes out of %d (seq=%"SCNu16", step=%"SCNu16", ts=%"SCNu64", time=%"SCNu64"s)\n", bytes, tmp->len, tmp->seq, diff, tmp->ts, (tmp->ts-list->ts)/8000); /* Decode and save to wav */ - AVPacket avpacket; - av_init_packet(&avpacket); - avpacket.data = (uint8_t *)buffer; - avpacket.size = bytes; + AVPacket *avpacket = av_packet_alloc(); @lu-zero <https://github.com/lu-zero> any reason why this allocation is inside the loop? I see for all other codecs you just do it once before the while, and just unref/initialize the packet at each loop, while for G.722 this is happening in the while instead. Is it because we're actually decoding something here, rather than just saving to a container? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2665 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAB2LJHTM4Z2K3KGI6SIZ7DTPNYBZANCNFSM4454KPIQ> .

lminiero · 2021-05-25T10:41:58Z

postprocessing/pp-opus.c

 				if(pos >= nextPos) {
-					JANUS_LOG(LOG_WARN, "[SKIP] pos: %06" SCNu64 ", skipping remaining silence\n", pos);
+					JANUS_LOG(LOG_WARN, "[SKIP] pos: %06" SCNu64 ", skipping remaining silence\n", pos / 48 / 20 + 1);


Mh looking at these changes again (with fresh eyes from @atoppi) there seems to be something broken in the way pos is updated and handled now. I see how you decided to normalize both pos and nextPos, but before pos depended on i, while now it doesn't: since tmp never changes in the for loop, neither will the value of pos, making the checks broken. Considering pos is used as a basis for the av_rescale_q below, this would likely not insert all the silences we need, in case multiple packets were lost in a row.

Oh, right. Updated to increase the pos by 20 * 48.

atoppi · 2021-05-25T13:42:05Z

hi @lu-zero ,
gave this PR a run with an opus recording sample (attached).
This specific sample has many 1-packet gaps and a single large gap of consecutive packets.
The new post-processor correctly filled all the gaps, so everything looks fine.

Anyway I have observed some differences in the ffprobe output:

Input #0, ogg, from 'old-pp.opus':
  Duration: 00:00:34.98, start: 0.000000, bitrate: 39 kb/s
    Stream #0:0: Audio: opus, 48000 Hz, stereo, fltp
    Metadata:
      comment         : {"t": "a", "c": "opus", "s": 1621947438860716, "u": 1621947438863033}

Input #0, ogg, from 'new-pp.opus':
  Duration: 00:00:34.96, start: -0.020000, bitrate: 30 kb/s
    Stream #0:0: Audio: opus, 48000 Hz, stereo, fltp
    Metadata:
      comment         : {"t": "a", "c": "opus", "s": 1621947438860716, "u": 1621947438863033}
      encoder         : Lavf58.29.100

As you can see the opus file from the PR (new-pp.opus):

has a shorter duration (20 ms) but a negative start time that seems to "compensate"
has a lower bitrate (30kbps vs 39kbps), indeed the file size is smaller (130K vs 171K)

What is the reason of those differences?

lu-zero · 2021-05-25T15:46:02Z

lavf adds the pre-skip, and has a different muxing strategy apparently, the same amount of packets is stored though.

ffprobe -show_packets shows that all the packets are stored that, I can see using vimdiff.

atoppi · 2021-05-26T09:45:15Z

Thanks for clarifying, according to the rfc:

A 'pre-skip' field in the ID header (see Section 5.1) signals the
number of samples that SHOULD be skipped (decoded but discarded) at
the beginning of the stream

AFAIU that means that a player will basically skip (e.g. not play) the first 20 ms (1 packet) of the opus file, while the current pp-rec is generating a file that will play since the very first packet.
Still I don't understand if the two files are "synchronized", I mean if they start playing in the same point, have same duration and there is any "loss" of information in the new one (e.g. will that pre-skip packet played somehow? why drop it?).
Sorry for the questions, I'm not exactly a codec guy.

As of my other point

has a lower bitrate (30kbps vs 39kbps), indeed the file size is smaller (130K vs 171K)

do you have any comment ?

lu-zero · 2021-05-26T09:52:13Z

It is an estimate

    if (ic->pb && (filesize = avio_size(ic->pb)) > 0 && ic->duration > 0) {
        /* compute the bitrate */
        double bitrate = (double) filesize * 8.0 * AV_TIME_BASE /
                         (double) ic->duration;
        if (bitrate >= 0 && bitrate <= INT64_MAX)
            ic->bit_rate = bitrate;
    }

lminiero · 2021-05-26T10:35:57Z

@lu-zero I think what @atoppi is mentioning are two separate things, that may or may not be an issue, and that's what we're trying to figure out:

If the pre-skip is causing the first 20ms of audio to be ignored by players (and media processors), that's a problem, as that may cause some desync problems when we're going to use the resulting media file in other contexts, e.g., to mux it with the corresponding video file. Is this something that libavformat always does, or is there a way to disable it? Or is it related to the pts/dts generated by av_rescale_q, that may generate negative values at startup?
The average bitstream is not just a matter of how it's calculated, but seems to be a consequence of files that actually are of a different size: the same MJR file (with some packet losses to test the silence ingestion) was converted to a 170KB file by the current processor, while it was converted to a 130KB file by the new one. This means that there are 40KB unaccounted for, one way or another, and it's not clear if it's because of differences in how silence is ingested. We'd expect the size to be more or less the same, as while there may be differences in how the ogg file header is created when using libogg difrecly vs. libavformat, we're manually adding a static silent audio packet, and packets that weren't lost all are expected to have the same in both files as well. We suspected it might be because of some transcoding happening somewhere (which would indeed result in a different Opus bitstream with a different bitrate), but looking at the code it does seem to be just saving the data to the container as we did for video streams already. Do you know what may be causing such a big difference instead?

Thanks!

lu-zero · 2021-05-26T14:06:56Z

@lu-zero I think what @atoppi is mentioning are two separate things, that may or may not be an issue, and that's what we're trying to figure out:

1. If the pre-skip is causing the first 20ms of audio to be ignored by players (and media processors), that's a problem, as that may cause some desync problems when we're going to use the resulting media file in other contexts, e.g., to mux it with the corresponding video file. Is this something that libavformat always does, or is there a way to disable it? Or is it related to the pts/dts generated by `av_rescale_q`, that may generate negative values at startup?

I set explicitly the packet duration. The original code assumed 20ms packets, I made it explicit. Now the timestamps are identical.

2. The average bitstream is not just a matter of how it's calculated, but seems to be a consequence of files that actually are of a different size: the same MJR file (with some packet losses to test the silence ingestion) was converted to a 170KB file by the current processor, while it was converted to a 130KB file by the new one. This means that there are 40KB unaccounted for, one way or another, and it's not clear if it's because of differences in how silence is ingested. We'd expect the size to be more or less the same, as while there may be differences in how the ogg file header is created when using libogg difrecly vs. libavformat, we're manually adding a static silent audio packet, and packets that weren't lost all are expected to have the same in both files as well. We suspected it might be because of some transcoding happening somewhere (which would indeed result in a different Opus bitstream with a different bitrate), but looking at the code it does seem to be just saving the data to the container as we did for video streams already. Do you know what may be causing such a big difference instead?

It is the muxing overhead, you can use ffprobe -show_packets ${file} | grep size | cut -d '=' -f 2 | paste -sd+ | bc to check.

Your sample has 128947 bytes of actual data.

You can remux using ffmpeg -i old.opus -c copy new.opus and get

172K	old.opus
132K	new.opus

lminiero · 2021-05-26T14:18:32Z

Oh I wasn't aware of muxing overhead and that it could be so high, thanks for the clarification and for the helpful snippet!

I set explicitly the packet duration. The original code assumed 20ms packets, I made it explicit. Now the timestamps are identical.

Do you mean in a new commit? I don't see it yet.

lu-zero · 2021-05-26T14:35:34Z

The push happened a bit later than the message :)

atoppi · 2021-05-27T11:21:23Z

@lu-zero tried with another recording sample of my voice and got the following with the new pp-rec

[ogg @ 0x61b000000080] Timestamps are unset in a packet for stream 0. This is deprecated and will stop working in the future. Fix your code to set the timestamps properly
[ogg @ 0x61b000000080] Encoder did not produce proper pts, making some up.
audio-new.opus is 70993 bytes
Bye!

Current version does not show any warning.
The generated file seems ok, though.

lu-zero · 2021-05-27T11:39:42Z

Pull and try again please.

atoppi · 2021-05-27T13:25:49Z

thanks, now everything looks good to me 👍

lminiero · 2021-05-27T13:28:33Z

We can merge then! ✌️
Thanks again for your contribution @lu-zero! (and for your patience going through our silly questions 🤭 )

lu-zero · 2021-05-27T13:38:16Z

Thank you for the review :)

commit 9c9d335 Author: Lionel Nicolas <lionelnicolas@users.noreply.github.com> Date: Thu Jun 10 03:04:43 2021 -0400 Fix streaming plugin mutex unlock when disabling mountpoint (meetecho#2690) commit 2d83e96 Author: Yurii Cherniavskyi <yurii.cherniavskyi@gmail.com> Date: Mon Jun 7 16:02:41 2021 +0300 Fix SIP plugin unhold request docs typo (meetecho#2688) commit 2cd0118 Author: August Black <augustblack@gmail.com> Date: Mon Jun 7 01:10:49 2021 -0700 minor adjustment to the audiobridge docs (meetecho#2687) commit de2117e Author: nicolasduteil <nduteil@freedev.org> Date: Tue Jun 1 11:26:29 2021 +0200 fix: [janus_sip] Fix "call_id" property in "missed_call" events (meetecho#2679) commit 9eeeb38 Author: Alessandro Toppi <atoppi@meetecho.com> Date: Mon May 31 15:57:41 2021 +0200 Fix status vector parsing for incoming twcc feedbacks (resolves meetecho#2677). commit 8a25f6e Merge: d3b39b9 394fb48 Author: Alessandro Toppi <atoppi@meetecho.com> Date: Fri May 28 13:29:54 2021 +0200 Merge pull request meetecho#2675 from kmeyerhofer/actions/fix GH Actions, fix variable name commit d3b39b9 Author: Lorenzo Miniero <lminiero@gmail.com> Date: Fri May 28 11:09:30 2021 +0200 Fixed race condition in VideoRoom commit 394fb48 Author: Kurt Meyerhofer <k@kcmr.io> Date: Thu May 27 14:52:08 2021 -0600 Fixes variable name. commit b45cd37 Author: Lorenzo Miniero <lminiero@gmail.com> Date: Thu May 27 18:31:55 2021 +0200 Clarify that libnice 0.1.18 is recommended commit 5757a37 Author: Lorenzo Miniero <lminiero@gmail.com> Date: Thu May 27 17:08:17 2021 +0200 Spatial audio support in AudioBridge via stereo mixing (meetecho#2446) commit 161fe7a Author: Luca Barbato <luca.barbato@gmail.com> Date: Thu May 27 15:29:01 2021 +0200 Cleanup avformat-based preprocessors (meetecho#2665) commit 7b010cd Author: lucylu-star <78361868+lucylu-star@users.noreply.github.com> Date: Tue May 25 17:09:40 2021 +0800 Fixed broken simulcast support in VideoCall plugin (meetecho#2671) commit 4ae44a4 Author: nicolasduteil <nduteil@freedev.org> Date: Mon May 24 17:57:34 2021 +0200 feat: support for custom call-id in subscribe request + add 'call_id' property to subscribe & notify related events (meetecho#2664) commit 4294f20 Author: Lorenzo Miniero <lminiero@gmail.com> Date: Mon May 24 11:02:48 2021 +0200 Fixed missing macro when using pthread mutexes (fixes meetecho#2666) commit f22ab0d Author: Lorenzo Miniero <lminiero@gmail.com> Date: Wed May 19 12:03:32 2021 +0200 Fixed warning commit b3f3f17 Author: Alessandro Toppi <atoppi@meetecho.com> Date: Tue May 18 12:10:47 2021 +0200 Remove duplicated flag for fuzzing coverage. commit 4a7560c Author: nu774 <honeycomb77@gmail.com> Date: Fri May 14 00:26:36 2021 +0900 janus-pp-rec: support HEVC AP(aggregation packet) (meetecho#2662) commit 5db4be2 Author: Lorenzo Miniero <lminiero@gmail.com> Date: Wed May 12 17:43:43 2021 +0200 Fixed out of bounds array access commit 69f56f4 Author: nicolasduteil <nduteil@freedev.org> Date: Tue May 11 14:36:22 2021 +0200 feat: support for SUBSCRIBE expiry (Expires header) in sip plugin (meetecho#2661) commit b047ccf Author: Lorenzo Miniero <lminiero@gmail.com> Date: Mon May 10 09:33:27 2021 +0200 Fixed types commit f8e8c5e Author: Chris Wiggins <chris@wiggins.nz> Date: Mon May 10 19:26:45 2021 +1200 RabbitMQ Transport Reconnect Logic (meetecho#2651) commit 280e8e4 Author: Lorenzo Miniero <lminiero@gmail.com> Date: Fri May 7 12:54:30 2021 +0200 Add per-participant recording options in AudioBridge to join API as well

lu-zero added 3 commits May 17, 2021 10:40

Factor out the avformat macro compat

4369055

Do not use av_init_packet

f156cc1

Do not allocate an encoder while muxing to webm

aa3c532

There is no need to allocate an encoder to mux. Makes possible to use the postprocessor without having a vp8/vp9 encoder in avcodec.

lu-zero force-pushed the cleanup_avformat branch from 563c381 to aa3c532 Compare May 17, 2021 08:52

lu-zero force-pushed the cleanup_avformat branch from fd4a541 to 26db9ba Compare May 21, 2021 18:45

lminiero reviewed May 24, 2021

View reviewed changes

lu-zero added 5 commits May 24, 2021 11:54

Factor out the video stream setup

e10142f

Remove a spurious copy

b995c6b

Factor out the context setup

50b7cdc

Move the common code in a stand alone file

739ebf0

Simplify the av1 postprocessor

c85b6c4

lu-zero force-pushed the cleanup_avformat branch from 26db9ba to c7e8c34 Compare May 24, 2021 09:55

lminiero reviewed May 24, 2021

View reviewed changes

lu-zero force-pushed the cleanup_avformat branch from c7e8c34 to cab9e5b Compare May 24, 2021 13:33

lminiero reviewed May 25, 2021

View reviewed changes

lu-zero force-pushed the cleanup_avformat branch 2 times, most recently from 7f0a9ed to cd57e6c Compare May 25, 2021 10:53

lu-zero force-pushed the cleanup_avformat branch from cd57e6c to 329d692 Compare May 26, 2021 14:29

Use libavformat instead of libogg for opus

f636ff2

lu-zero force-pushed the cleanup_avformat branch from 329d692 to f636ff2 Compare May 27, 2021 11:39

lminiero merged commit 161fe7a into meetecho:master May 27, 2021

lu-zero deleted the cleanup_avformat branch May 27, 2021 13:37

lminiero mentioned this pull request Jun 1, 2021

Additional target formats for some recorded codecs (fixes #2658) #2680

Merged

atoppi mentioned this pull request Jun 1, 2021

postprocessing #2678

Closed


		/* WebM output */
		fctx = janus_pp_create_avformatcontext("ogg", metadata, destination);

Cleanup avformat-based preprocessors #2665

Cleanup avformat-based preprocessors #2665

Conversation

lu-zero commented May 15, 2021

januscla commented May 15, 2021

lminiero commented May 17, 2021

lu-zero commented May 17, 2021 • edited

lminiero commented May 17, 2021

lu-zero commented May 19, 2021

lminiero commented May 19, 2021

lminiero commented May 21, 2021

lu-zero commented May 21, 2021 • edited

lminiero commented May 21, 2021

lminiero left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lu-zero commented May 24, 2021

Choose a reason for hiding this comment

lminiero commented May 24, 2021

lu-zero commented May 24, 2021

lminiero commented May 24, 2021

lu-zero commented May 24, 2021

lminiero commented May 24, 2021

Choose a reason for hiding this comment

lu-zero commented May 25, 2021 via email

Choose a reason for hiding this comment

Choose a reason for hiding this comment

atoppi commented May 25, 2021

lu-zero commented May 25, 2021 • edited

atoppi commented May 26, 2021

lu-zero commented May 26, 2021

lminiero commented May 26, 2021

lu-zero commented May 26, 2021

lminiero commented May 26, 2021

lu-zero commented May 26, 2021

atoppi commented May 27, 2021

lu-zero commented May 27, 2021

atoppi commented May 27, 2021

lminiero commented May 27, 2021

lu-zero commented May 27, 2021

lu-zero commented May 17, 2021 •

edited

lu-zero commented May 21, 2021 •

edited

lu-zero commented May 25, 2021 •

edited