[Vorbis] [EXT] Re: oggenc argument to make the stream not cut in the middle

[Jan Stary]

First of all, keep the replies on the mailing list.
Are you subscribed to the mailing list?

On Aug 24 09:44:56, nguyenquocthao00 at gmail.com wrote:
> The file ogg is the same, I know that.

So oggenc must have produced exactly the same audio.
So it must be specific to the "live" playing.

> And the problem is not from curl,
> curl is only an example to replicate my problem.
> If I download the file and
> use cat instead of curl, it's the same

Aaaargh, so why even bring curl into that?
I'm working wth a local copy of dataproc.wav now, OK?

( cat dataproc.wav && sleep 0 ) | oggenc - | play - 
( cat dataproc.wav && sleep 4 ) | oggenc - | play - 
( cat dataproc.wav && sleep 8 ) | oggenc - | play - 

> All this has the same audio,

Exactly, it just waits 4 resp. 8 seconds before encoding and playing.
No surprise there.

> the difference is that there is silence when speaking the sentence,
> making the sentence not spoken smoothly

I cannot hear any such silence. The sentence sounds the same.

> Using another example:
> In my computer, using this command
> ( curl https://storage.googleapis.com/maika-ai-ext/test/test.wav && sleep
> 8) | oggenc - | play -

This test.wav is not a valid wav file:
play FAIL formats: can't open input file `test.wav': invalid chunk ID found
ffmpeg can read it, for instance, but can you please give use a valid file?

> When speaking `for content marketing`, the word `content` is split in the
> middle, only speak `con` then delay, then speak `tent marketing`

OK, now I see what you mean.  Testing with vorbis-tools-1.4.2
and sox-14.4.2pl20210509p1 on OpenBSD 7.3:

( cat test.wav && sleep 0 ) | oggenc - | play - 
( cat test.wav && sleep 4 ) | oggenc - | play - 
( cat test.wav && sleep 8 ) | oggenc - | play - 

For me, the "delay" happens between "specificaly" and "designed".
Note that this does _not_ happen with the previous dataproc.wav file.

> If the is no sleep, sleep 0, or play directly without using oggenc -, the
> sentence is spoken smoothly
> ( curl https://storage.googleapis.com/maika-ai-ext/test/test.wav && sleep
> 8) | oggenc - | play -
> ( curl https://storage.googleapis.com/maika-ai-ext/test/test.wav && sleep
> 8) | ffplay -
> 
> The problem is that oggenc need the full data before writing to stdout, so
> if the stdin needs to wait for sometime before closing, some data will be
> held in buffer. The problem is the delay between sending data / receiving
> data.

OK, so you know the sleep makes the buffers wait.
And it only occurs with "live" playback of the encoded ogg,
not when writing with oggenc -o, where it produces the same
output in all three cases.

Yes, live playback is sensitive to timing and buffering.
So what is your question?

> https://lists.xiph.org/pipermail/vorbis/2002-August/021175.html
> https://lists.xiph.org/pipermail/vorbis/2005-July/025835.html
> If you still cannot replicate the problem, please let me know

I can now; it also happens with ogg123 in place of play,
although at a different point.

I don't know what you want to do about it;
oggenc buffers its input and output,
and that's what you are messing with.

	Jan


> > On Aug 23 08:51:15, hans at stare.cz wrote:
> > > For reference, I attach the two outputs:
> > >
> > > ( curl https://storage.googleapis.com/maika-ai-ext/test/dataproc.wav \
> > > && sleep 0 | oggenc -o sleep-0-oggenc.ogg
> > >
> > > ( curl https://storage.googleapis.com/maika-ai-ext/test/dataproc.wav \
> > > && sleep 4 | oggenc -o sleep-4-oggenc.ogg
> > >
> > > Listening to both, I hear exactly the same.
> > >
> > >       Jan