[asterisk-users] Silk for Free

[Steve Underwood]

Wilton Helm wrote:
> >12kHz isn't really enough for high quality voice, and the extra bit
> >rate needed to push the bandwidth to 15kHz is small. Also, a deep man's
> >voice looses something when you cut off at 70Hz.
>  
> I'm not sure that this isn't stretching things a bit.  There are no 
> handsets or headsets (AFAIK) that can do justice to 50 KHz and 
> probably most speakers attached to a PC can't.  Likewise, while a deep 
> male voice can go below 70 Hz, few transducers can do justice to those 
> frequencies, either.  I don't think the attempt is to reproduce a 
> symphony.  The extra bandwidth (even if it is minor) would be hard to 
> justify if one needed $500 speakers to benefit from it.  While a 
> number of people might be able to tell the difference in an A B 
> comparison, I suspect few would notice it without direct comparison.  
> I also suspect Skype is correct in that the majority of people, 
> listening to it on typical hardware would like additional low 
> frequencies less than without because of things like distortion in the 
> transducer.
So, your approach is to base a high quality codec around the crappiest 
transducers it will ever meet? :-\ If you want to define standards that 
stick around, you try to make them work beyond the junk you see today. 
Making a wideband handset isn't hard. Its a couple of dollars to make a 
really good one, not $500. People just haven't bothered, because 
wideband telephony hasn't really existed to make use of it. A 20 cent 
handset saves them money, but $2 isn't a lot to pay for a handset in a 
$100 phone.
>  
> Getting the bandwidth above 3 KHz at the top will improve 
> intelligibility, but somewhere between 5 and 10 KHz that reaches a 
> point of diminishing returns.  Likewise, extending the low end below 
> 300 Hz will help naturalness, but that also reaches diminishing 
> returns somewhere around 100 Hz unless all the pieces are very high 
> quality (from the mic to the speaker).  It seems to me that they have 
> exceeded those realities by a comfortable margin, which is generally 
> what good engineering is all about.
Skype say they are sampling at 24k for their widest bandwidth, so they 
won't really get 12kHz. More like 11kHz, I guess. Extending  voice up to 
7kHz pretty much fixes all the intelligibility problems inherent in 
normal narrowband telephony. However, the difference in sound quality 
between 7kHz bandwidth and 11kHz is big, and going up to 15kHz adds 
significantly more. It takes very few extra bits to extend the frequency 
range. Doubling the bandwidth does not double the bit rate with modern 
compressed codecs.

Good engineering of standards is about building them for the future. 
Cutting off the bass at 70Hz is far less of a limitation than cutting 
off the high end at 11kHz, but why do it in the codec? Why not leave the 
transducers and their amps to do what they can?

Regards,
Steve