Re: [codec] Conformance with unusual sample rates

[Gregory Maxwell <[e-mail]>]

John Ridges [[e-mail]] wrote:
> Since the codec conformance is a hot topic right now, I thought I would
> ask a (probably dumb) question. Is it possible to create a conforming
> (and thus unencumbered) Opus codec from the reference implementation
> that can use a sample rate of, say, 44100 Hz? If so, what are the rules
> that would guarantee conformance? Thanks,

The _correct way_ to use Opus with 44100 Hz is to use a resampler. 

This is, for example, how the opus-tools package uses the reference
library to support any input and output sampling rates the user desires,
transparently and painlessly. (https://git.xiph.org/?p=users/jm/opus-tools.git)

Personally, I think that it's most accurate to think of Opus as a sample-rateless
format. Any rate can go into the encoder, you get packets out at 2.5-120ms 
intervals, and any rate can come out of the decoder.  For particular rates
(8000,12000,16000,24000,48000) the conversion from opus to audio is extra
efficient so those are the ones that the reference library supports directly.
(But not other rates because platform specific optimized resampling would
do better)

The computational cost of a resampler is not significant compared to
the codec, there is no quality impact from a reasonable resampler. Any
other option and you'd suffer from loss of coding efficiency, loss of
interoperability, lost of on the fly mode agility, etc.