[FFmpeg-devel] Fun with aevalsrc + audio testing

Nicolas George nicolas.george at normalesup.org
Fri May 18 16:28:39 CEST 2012

Le decadi 30 floréal, an CCXX, Clément Bœsch a écrit :
> I had a look to how we could test a bit the audio filter chain with
> aevalsrc as suggested by Stefano now that we have -af.

Now that we have -af, we can use files as input, that is precisely the

> So I had fun with aevalsrc and tried to "compose" some random sounds. Here
> is the result that could be used for 1 of the 6 channels of that potential
> input stream:

That is quite interesting. But there are two issues with aevalsrc: first, it
is rather slow; second, and most important: it is not bit-exact at all.

IMHO, there is need for a source "atestsrc", that produces varied (but
deterministic) patterns on several channels. These patterns may include:

- piano-like pure notes, like the one you produced;

- guitar-like notes with attack and noise;

- pitch slides like parts of audiogen;

- ... basic voice synthesis anyone?

If different kinds of sounds (or pitch) are placed on different channels,
recognizing which channel is which by listening is possible.

As for bit-exactness, some time ago I wrote a integer-only, reasonably fast
and accurate sine table generator. I can dig it up if anyone believes it


  Nicolas George
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20120518/066f9b9b/attachment.asc>

More information about the ffmpeg-devel mailing list