[Sigia-l] Serious Discussion of IA Research?

Boniface Lau boniface_lau at compuserve.com
Sat Dec 4 20:58:29 EST 2004


> From: Andrew Boyd
>  
> Voice carried on an eight-bit channel is sometimes woefully
> distorted, 

You don't even need 8 bits to encode a human speech sample with
comparable analog quality. Plain old pulse code modulation can do that
in 7 bits. The distortion that you referred to is often due to overly
aggressive sharing of transmission medium.


> but we get by on it, rather than it being of 32 bit 'stereo
> quality', by use of the adaptive code in the phones

Adaptive encoding requires less sampling data to be transmitted and
therefore allows more sharing of transmission medium.


> and the adaptive pattern-matching inside our heads.

Our mind smoothes over the gaps in what it heard.


[...]
> As a general rule, we don't get the best, but an acceptable level of
> service/bandwidth/quality, because it is not profitable to do the
> best.

Yes. By understanding entropy, we have more choices of allowable
distortion, ranging from a lot to none.


> I think that it is important to recognise the difference between 'the 
> best we can provide with what we have' and 'the best possible service 
> available'. This applies to the whole field of human endevour.

Indeed. That extra last mile can be prohibitively expensive to cross.


Boniface




More information about the Sigia-l mailing list