[nylug-talk] Sound Analysis and CPU power

[Joseph Gaffney]

On Apr 2, 2007, at 3:24 PM, alex at pilosoft.com wrote:

> On Mon, 2 Apr 2007, Ruben Safir wrote:
>
>>  Meanwhile, this genetleman was saying that the sound analysis  
>> programming
>>  he uses runs into problems with realtime usage.  At first I  
>> thought this
>>  was another case of just bad software design.  But as it was  
>> explained
>>  it might be very well the case that the analysis of sound can be  
>> a tough
>>  nut to crack even for a modern dual core intel chip.  Why is it that
>>  we still don't have fundementally sound software for voice  
>> recognition and
>>  dictation will all this power under the hood?  Is it that we  
>> reach hardware
>>  threasholds, or is it just poor software design and not enough  
>> qualified
>>  engineers interested in the problem?
> DSPs are hard. Let's go shopping.
>
> Yes, current processors are pretty darn fast - what could only be  
> done on
> DSP 5 years ago can be done on a commodity processor today. However,
> current DSPs are even faster than 5 year old ones. So, for example,  
> sure,
> you can do (random example, numbers pulled out of my butt) 20 channels
> of echo cancellation on a P4 that cost 200$, but on a 200$ DSP you  
> can now
> do 500 channels of echo cancellation.
>
> Right tool for the right job.
>
> -alex

Nail right on the head.

For example, Shure's new ATC system is essentially 16 DSP's with an  
optional analog phone add (otherwise its just your usual fare of  
gozintas and gozoutas).  Try as I might, I could not, for the life of  
me, overload this monster.

Voice recognition? Think appliance, not application.

-Joe