[Spice-devel] RFC on sound codec refactoring

[Jeremy White]

On 10/08/2013 08:44 AM, Jeremy White wrote:
>> However, I can see the record path has a fixed 44100 value, which is
>> probably bogus today for != 44100 recording?
> 
> Yah; there is a fair amount of hard coding (although I'm probably
> overly biased by my time in the old client :-/).
> 
> And, afaik, there's not a lot of testing done at anything other than 44100,
> so we may find corner cases yet.

For the record, I've done some probing into the effects of the hard
coded 44100, and I believe it does no harm, particularly if we run at 48000.

The hard coding only affects the size of the compressed celt buffer;
it's computed to be this big:

#define FRAME_SIZE 256
#define CELT_BIT_RATE (64 * 1024)
#define CELT_COMPRESSED_FRAME_BYTES (FRAME_SIZE * CELT_BIT_RATE / 44100 / 8)

which ends up being 47 :-/.  Basically, the size of the buffer dictates
how much compression to perform.  If the frequency goes higher, that
logic actually allows for a smaller buffer; i.e. we can compress the
same number of samples a bit more.  But in our 'worry' case, all that
happens is we continue to use 47 bytes instead of 43, and do a slightly
less efficient compression job.

So I think we'd only have troubles if we ran well below 44100, and then
only for old clients connecting to new servers, and only for record mode.

So, um, <small>nevermind</small>.

Cheers,

Jeremy