Stereo PCM

[elshnkhll]

Hi,
Thanks for a great library!
I am using OPUS decoder a lot.
Having a small problem though. I would like to have left and right channel data in WAV (interleaved/merged) format like:
(L,1),(R,1),(L,2),(R,2),...,(L,n),(R,n)
Is it possible? Merging them in browser JS 'for' loop is little expensive.
With current structure player must deal with splitters/mergers etc.
May be there are simpler ways of playing it i am not aware of.
Thanks in advance.

[japj]

How are you encoding the opus file?

I think I am running into something similar.

I was looking into coupled_streams as part of the opus multistream api.
But I think even when you create an opus file with several coupled_streams, I think the (ogg) opus decoder decodes into (mono) channels.

I have a custom ffmpeg that now creates a mapping_family 255 opus file with 4 coupled streams, but when decoding I get all the 8 separate (mono) channels instead of 4 stereo streams.

e.g. I have:

Processing file "data/shower-the-people/final-coupled-output.opus"...

New logical stream (#1, serial: adeb0cdf): type opus
Encoded with Lavf61.7.100
User comments section follows...
        encoder=Lavc61.19.101 libopus
        DESCRIPTION=Channel order: Sop, Alt, Ten, Bass
Opus stream 1:
        Pre-skip: 312
        Playback gain: 0 dB
        Channels: 8
        Original sample rate: 48000 Hz
        Streams: 4, Coupled: 4
        Channel Mapping Family: 255 Map: [0, 1, 2, 3, 4, 5, 6, 7]
        Packet duration:   20.0ms (max),   20.0ms (avg),   20.0ms (min)
        Page duration:   1000.0ms (max),  998.3ms (avg),  620.0ms (min)
        Total data length: 11659105 bytes (overhead: 0.493%)
        Playback length: 3m:44.597s
        Average bitrate: 415.3 kbit/s, w/o overhead: 413.2 kbit/s
Logical stream 1 ended

Background: I am working on a learning track encoder/playback website that uses the opus format to encode the different parts (Soprano, Alt, Tenor, Bass). Each of these parts is a stereo track (with 3 sung versions mixed Left, Center, Right).

In the end I want to control the stereo track for each part, but it seems I get all the mono channels separately instead.

[japj]

I tried something else, generating an ogg file with multiple (stereo) streams.

ffmpeg -stats -y \
-i ./data/shower-the-people/S.wav \
-i ./data/shower-the-people/A.wav \
-i ./data/shower-the-people/T.wav \
-i ./data/shower-the-people/B.wav \
-i ./data/shower-the-people/S-P.wav \
-i ./data/shower-the-people/A-P.wav \
-i ./data/shower-the-people/T-P.wav \
-i ./data/shower-the-people/B-P.wav \
-filter_complex "\
[0:a][4:a]amerge=inputs=2[s_stereo]; \
[1:a][5:a]amerge=inputs=2[a_stereo]; \
[2:a][6:a]amerge=inputs=2[t_stereo]; \
[3:a][7:a]amerge=inputs=2[b_stereo]" \
-map "[s_stereo]" -map "[a_stereo]" -map "[t_stereo]" -map "[b_stereo]" \
-c:a libopus -b:a 96k \
-mapping_family 1 \
-metadata:s:a:0 comment="Soprano Stereo (S + S-P)" \
-metadata:s:a:1 comment="Alto Stereo (A + A-P)" \
-metadata:s:a:2 comment="Tenor Stereo (T + T-P)" \
-metadata:s:a:3 comment="Bass Stereo (B + B-P)" \
./data/shower-the-people/minimal-output-stereo-pairs.opus

This generates 4 stereo streams with the vocal left and piano right.
And then creates one ogg file with those 4 streams

Output #0, ogg, to './data/shower-the-people/minimal-output-stereo-pairs.opus':
  Metadata:
    encoder         : Lavf61.7.100
  Stream #0:0: Audio: opus, 48000 Hz, stereo, s16, 96 kb/s (default)
      Metadata:
        DESCRIPTION     : Soprano Stereo (S + S-P)
        encoder         : Lavc61.19.101 libopus
  Stream #0:1: Audio: opus, 48000 Hz, stereo, s16, 96 kb/s
      Metadata:
        DESCRIPTION     : Alto Stereo (A + A-P)
        encoder         : Lavc61.19.101 libopus
  Stream #0:2: Audio: opus, 48000 Hz, stereo, s16, 96 kb/s
      Metadata:
        DESCRIPTION     : Tenor Stereo (T + T-P)
        encoder         : Lavc61.19.101 libopus
  Stream #0:3: Audio: opus, 48000 Hz, stereo, s16, 96 kb/s
      Metadata:
        DESCRIPTION     : Bass Stereo (B + B-P)
        encoder         : Lavc61.19.101 libopus

but it seems the playback is not working correctly
instead of 4 stereo streams played back, I get chunked sequential parts

expected, SATB1, SATB2, SATB3, SATB4
I get, S1A1T1B1, S2A2T2B2, S3A3T3B3, S4A4T4B4

where 1,2,3,4 is opus packet size (I think)

I am using ogg-opus-decoder, could it be that it expects only one ogg stream?

[eshaz]

Hi, Thanks for a great library! I am using OPUS decoder a lot. Having a small problem though. I would like to have left and right channel data in WAV (interleaved/merged) format like: (L,1),(R,1),(L,2),(R,2),...,(L,n),(R,n) Is it possible? Merging them in browser JS 'for' loop is little expensive. With current structure player must deal with splitters/mergers etc. May be there are simpler ways of playing it i am not aware of. Thanks in advance.

Currently this library returns data that's structured to be played using the WebAudio api, example. To play in the browser you can directly send the decoded audio to that API and avoid interleaving or changing the format. You should only need to restructure to interleaved if you're downloading or saving the decoded audio. Here's how the wav file is generated when downloading from the demo page.

[eshaz]

but it seems the playback is not working correctly instead of 4 stereo streams played back, I get chunked sequential parts

expected, SATB1, SATB2, SATB3, SATB4 I get, S1A1T1B1, S2A2T2B2, S3A3T3B3, S4A4T4B4

where 1,2,3,4 is opus packet size (I think)

I am using ogg-opus-decoder, could it be that it expects only one ogg stream?

@japj There's currently no support for structuring the decoded audio into tracks using ogg streams. I would avoid using those. I think what you did in FFmpeg would create 4 distinct Opus files / streams that are muxed into a single Ogg file.

In the end I want to control the stereo track for each part, but it seems I get all the mono channels separately instead

Its my understanding that the coupled streams in Opus are meant to group stereo tracks, or tracks that share similar data, for better compression. I don't think it's meant to be used as metadata to know that the channels should be grouped.

It's probably best if you create a multitrack Opus file, mux it into a single Ogg stream, and then convey the track information to your application in the Ogg metadata or as a separate file. This library currently doesn't read Ogg metadata, so if you choose that route, you'd need another library for that.