When playing WEBRTC, sometimes there is noise in the audio.

[nowgoing]

Description'

Please ensure that the markdown structure is maintained.

When playing the same video streaming source using WEBRTC, there is sometimes noise in the audio, but there is no noise when pulling the stream using RTMP.

1. SRS version: 4.0.37
2. The log of SRS is as follows:

1. The configuration of SRS is as follows:

Use the configuration file conf/rtc.conf.

Replay

How to replay bug?

Steps to replay the bug:

1. ffmpeg rtmp streaming
2. webrtc playback, check for noise in the audio
3. If there is no noise, stop streaming and repeat steps 1-2-3

Expected Behavior:

Expecting webrtc playback to have no audio noise.

TRANS_BY_GPT3

[PieerePi]

See PR #1969.

TRANS_BY_GPT3

[nowgoing]

See PR #1969.

> Test noise solution, thank you

TRANS_BY_GPT3

[PieerePi]

I saw that you said you made some changes and solved something before. Is it like what you said, there was no initialization or something like that?

TRANS_BY_GPT3

[nowgoing]

I saw that you mentioned before that you made some changes and resolved something. Was it like you said, without any initialization issues?

No, now we are using your PR.

TRANS_BY_GPT3

[winlinvip]

I have also encountered noise, which means not being able to hear normal sounds, but only chaotic noise.
Restart the server and then refresh the page, and the noise will be gone. I don't know where this noise is coming from.

TRANS_BY_GPT3

[PieerePi]

@winlinvip, please take a look at PR #1969.

TRANS_BY_GPT3

[nowgoing]

@winlinvip please take a look at PR #1969.
Make sure to maintain the markdown structure.

Using OBS (setting the media source as an mp4 file) for RTMP streaming, setting the audio for streaming as stereo, and playing the stream with WebRTC; the chance of noise reproduction is very high.

@PieerePi
In the old code, within the function SrsAudioDecoder::decode...

for (int i = 0; i < frame_->nb_samples; i++) {
     if (size + pcm_size * codec_ctx_->channels <= max) {
           memcpy(buf + size,frame_->data[0] + pcm_size*codec_ctx_->channels * i, pcm_size * codec_ctx_->channels);      
           size += pcm_size * codec_ctx_->channels;
      }
}

Copying each sample one by one is meaningful to avoid losing too much sample data when it exceeds the maximum value.
So I made some changes to the old code, as follows...

for (int ch = 0; ch < codec_ctx_->channels; ch++) {
       for (int i = 0; i < frame_->nb_samples; i++) {
               if (size + pcm_size  <= max) {
                    memcpy(buf + size, frame_->data[ch] + pcm_size * i, pcm_size);
                    size += pcm_size;
                }
         }
 }

There was also no noise in the test.

TRANS_BY_GPT3

[PieerePi]

Copying each sample one by one is meaningful to avoid discarding too much sample data when it exceeds the maximum limit.

@nowgoing
If it were me, I would choose to discard the entire channel of data. :-)

TRANS_BY_GPT3

[winlinvip]

👍

[winlinvip]

Fixed in 4.0.81 #2106 (comment)