[CELT-dev] CELT 0.11.1 is out

[Gregory Maxwell]

On Tue, Feb 15, 2011 at 1:55 PM, Timothy B. Terriberry
<tterribe at xiph.org> wrote:
>> The API should also be close to the final API. Compared to 0.10.x there
>> are also small quality improvements, mostly at high bit-rate.
>
> Jean-Marc tends to forget how long ago 0.10.x was... there were actually
> some substantial improvements at low rates, including his new
> "anti-collapse" feature which prevents total energy collapse right
> before transients (the anti-pre-echo described in Monty's CELT demo:
> http://people.xiph.org/~xiphmont/demo/celt/demo.html) for 10 and 20ms
> frames. A number of other minor changes also accumulated into some
> substantial quality gains at low rates, larger even than the ones at
> high rates, especially for smaller frame sizes.
>
> The encoder should also no longer fail when it runs out of room for the
> coarse energy (which, being variable-sized, was hard for a caller to
> predict), or any other coded values. This means it should now be safe to
> run CELT at unreasonably low rates (down to 2 bytes/frame).


This graph shows the average PEAQ ODG change between 0.11.1 and 0.10
for 48KHz, Mono, 10ms frames, as a function of bitrate over a fairly
diverse collection of inputs:

https://people.xiph.org/~greg/celt/0.11.1.480.png

PEAQ ODG isn't as good as real listening tests but these results
support the expectation of a small across the board improvement.