[Speex-dev] Sampling Rate

[Jean-Marc Valin]

Alain M. a écrit :
> What would be speex configuration recomended for Telefone/Voip quality
> voice? With a quality just a little better/similar then G.729? or GSM?

I generally recommend wideband if possible. Otherwise, I recommend 15
kbps, which is much better than G.729 or GSM. The reason for that is
that you're already pushing 16 kbps of (RTP+UDP+IP) headers around, so
it's a bit silly to try and compress the audio down to 8 kbps (unless
you have to because 15 kbps won't fit).

> is there a comparison chart somewhere, but telephone quality oriented?

A while ago, I ran PESQ on Speex and other codecs. The results for NB
and WB are at http://people.xiph.org/~jm/aes_nb.png and
http://people.xiph.org/~jm/aes_wb.png but SHOULD BE TAKEN WITH A GRAIN
OF SALT because PESQ is particularly unreliable for comparing codecs.

	Jean-Marc

> Thanks,
> Alain
> 
> Tom Grandgent escreveu:
>> Kirk,
>>
>> Speex was designed for 8kHz, 16kHz, and 32kHz sample rates.  If you
>> don't use one of these sample rates, you'll be messing up important
>> assumptions deep within the codec.  Why these sample rates?  It's
>> telecommunications tradition, rather than PC audio tradition.
>>
>> If you want an efficient and high quality format for voice chat, try
>> 16kHz with VBR quality 6.  You should see around a 10:1 compression
>> ratio when someone is talking.  That is, around 25kbps would be a
>> rough peak using these settings.
>>
>> If that's too much bandwidth for you, you can cut it by almost half
>> using VBR quality 2.  (The loss of quality will be noticible to most
>> people using headsets.  It is less noticible when using speakers.) 
>> For further bandwidth savings you could use 8kHz, but it's too much of
>> a quality hit to be worth it in my opinion.
>>
>> Tom
>>
>> <khaynes at kirkgames.com> wrote:
>>> Hi,
>>>
>>> I'm attempting to use speex for voice chat in my game.
>>>
>>> I'm capturing audio from the mic using OpenAL 1.1 in 16 bit mono.
>>> What sample rate is recommended for capturing from the mic? Which
>>> would be better: 11025 samples per second, or 22050 samples per
>>> second of 16 bit mono audio with respect to then encoding/decoding it
>>> with speex? I'm using 11025 to reduce bandwidth at the moment to
>>> capture.
>>>
>>> Given your answer to the above question, what is the most efficient
>>> frame size for encoding the samples to speex frames? The sample code
>>> uses 160 samples (320 bytes) of 16 bit audio samples, which are then
>>> encoded to 200 bytes of encoded data. Is a 30% reduction in data
>>> pretty standard? How do I use speex most efficiently?
>>>
>>> Kirk
>>>
>> _______________________________________________
>>> Speex-dev mailing list
>>> Speex-dev at xiph.org
>>> http://lists.xiph.org/mailman/listinfo/speex-dev
>>
>> _______________________________________________
>> Speex-dev mailing list
>> Speex-dev at xiph.org
>> http://lists.xiph.org/mailman/listinfo/speex-dev
>>
>>
> 
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev
> 
>