[Asterisk-Users] Audio quality

[Scott Lamb]

I'm a new Asterisk user. I've managed to set it up to do everything I  
want except sound good. Currently, Asterisk sounds considerably worse  
than my cell phone. I know VOIP can be _better_ than my cell phone,  
because I've heard Skype do it. (Using 32k iLBC, I believe.)

I did an experiment with audio quality:

1) I made a recording which was pretty good. I used an iSight  
microphone and recorded at its native 48k sampling rate in Audacity.  
Then I trimmed it and ran their noise reduction filter.

2) I transformed it into a bunch of other formats with this script:

     for base in "$@"; do
         for rate in 8000 16000 32000; do
             sox "$base"-48000.wav -r $rate "$base"-$rate.wav  
resample -ql
             speexenc --vad --dtx --rate $rate "$base"-$rate.wav  
"$base"-$rate.spx
             speexdec "$base"-$rate.spx "$base"-$rate-fromspx.wav
         done
         sox "$base"-8000.wav "$base"-8000.gsm
     done

3) I listened to them all with Quicktime Player.

 From that I determined that a WAV with 32k sampling rate sounds good  
to me, but I'm not happy with 8k or 16k. Since adding lossy  
compression is only going to make things worse, that rules out most  
of Asterisk's codecs.

The 32k (ultra-wideband) speex with default settings sounds good, and  
it's a free codec. I'm not real concerned about bandwidth use, but it  
does pretty well there, too.

I understand that Skype uses ultra-wideband iLBC; it's audio quality  
is also good. But I understand there are patent problems. (Are  
licenses even available? How much do they cost?)

I've got three problems for actually using this though:

1) Can asterisk load 32k audio files? I see that the current CVS's  
format_wav.c can not. Is there another module that does? I might try  
modifying the WAV loader. Are there assumptions of 8k throughout  
Asterisk, or is this pretty isolated to that file?

2) Are there higher quality versions of the Asterisk sounds? They're  
in a nice, professional voice, but 8K GSM just sounds awful. If the  
source encodings are available, I could resample them into whatever  
else in a similar fashion to what I've done above.

3) Are there clients that support the 32k sampling rate?

a) I'm particularly interested in softphones for OS X. The ones I see  
are X-Lite, SJphone, and iaxComm. X-Lite has some weird distortion  
problems for me with even 8k ulaw (<http://support.xten.net/ 
viewtopic.php?t=3626>). SJphone's Speex support is weird - they have  
8.0k and 15.2 k (?). iaxComm has the settings I want, but they sound  
awful. I'm not giving it a fair test, though; see #1.

(None of these have a decent user interface either, but one problem  
at a time...)

b) Any hardware phones?

Regards,
Scott

-- 
Scott Lamb <http://www.slamb.org/>