🚀 go-pugleaf

In article <1992Nov12.180027.5230@bnr.ca> mwandel@bnr.ca (Markus Wandel) writes:
>In article <1992Nov12.023508.13285@tamsun.tamu.edu> rahhelp@rigel.tamu.edu (Ronald A. Huizar) writes:
>>
>>One problem that I am already see is that the microcontroller I was
>>planning to use the MC68HC811E2 can only address about 64K of external
>>memory, and we are planning/contemplating storing the message
>>digitally. I have not had much DSP, my partner has had more than I,
>>but I figure that for the telephone we will need to sample at least
>>6000 Hz (since the phone BW is 30-3000, you remember Nyquist don't
>>you). For a 1 minute (kinda long I know) message that will require
>>about 3 Meg of memeory. To much for the E2. So I know that the
>>microprocessors can handle the amount of memory I would need but I did
>>want to have build my own computer first.
>

Why not do some compression while you are getting the data? You can try one of
the data bandwidth reduction methods -- delta modulation etc should 
be pretty easy to do. I guess obtaining a model for short-term data may also
be possible (this is just a thought) computing the reflection coefficients etc
is pretty optimized now.

I have a "Phone-Mate" answering machine that does digital recording of
messages. It takes messages of total length about 35minutes. And this thing
costed me about 130 bucks. So, if you can compute the amount of memory (10MB!)
you will know that they are not just saving the message samples directly! The
reconstructed speech is a little rusty but it's alright. 

There is a TI Voice-Band Audio Processor that seems like a pretty good thing
to use for a project like this. It has stuff for preprocessing and
postprocessing (amplification) etc. built onto it.

This is just off the top of my head. My $0.02.

-- Girish Govind