Heh heh heh, this reminds me of back when those of us who wanted to
play or compose MOD files but were too broke to buy sound cards had
to hack the PC speaker to do it... and technically, that was a 1-bit
device!

If you want slightly better reproduction from 10-bit output without
any additional data space cost, you can try this:

Record 8 bits, but when reproducing, use a fast curve fit (such as a
travelling Bezier) to reproduce the low 2 bits.  You should use a
very small window for the fit.  Whether or not you will have
processing time to do this will depend on your sampling rate and chip.

Also remember that the lower your resolution, the less room you have
for error in optimizing for dynamic range, so what you really want to
be sure you do is record in 16 bit, normalize the volume (Cool Edit
or Sound Forge is suggested for this, though lots of programs will
do it) and then export it to your device.

One other weird qualitative thing that can sometimes help.  This should
never matter, but sometimes it seems to make a differrence.  Record in
16-bit STEREO, using two microphones if you're doing speech, THEN
subsample to 8-bit stereo, THEN mix down from stereo to mono.  I know,
I know-- but I'll be damned if sometimes it doesn't sound better.

Audio is weird.  Digital audio is that much weirder, because you can't
wave your hands as much. =3D]

Cheers,
   Matt

> -----Original Message-----
> From: A.J. Tufgar [mailto:tufgaraj@MUSS.CIS.MCMASTER.CA]
> Sent: Wednesday, October 23, 2002 7:58 AM
> To: PICLIST@MITVMA.MIT.EDU
> Subject: Re: [EE]: speech
>=20
>=20
> Try looking on Atmel's website for the AVR335 app note, it's got a
> really good method for reproducing speech.
>=20
> With this medthod I was able to use the '872 @ 20Mhz to obtain a 10bit
> PWM @ 19.5Khz, to produe speech much like everyone has been talking
> about.  I think the quality is amazing for what it costs to implement.
>  Try using a LMC6484 instead of the LM324.
>=20
> There are two disadvantages:
> One, I seem to get a popping or crackling if I set my volume=20
> to high on
> an external amp I'm not yet sure why.  It doesn't seem to be the amp
> though.  I think I'm hearing some of the PWM.  If I find out why I'll
> post it to the list.  Could be my LMC6484 used for filtering PWM.
>=20
> Second if you choose to use 10bit PWM, it's annoying to work=20
> with.  You
> must either encode to 8 (I would try recording the differential rather
> then absolute) or you could record the 10 bits and use 5 registers for
> 4 bits of info.  Your choice, but let's just say it's annoying to
> implement the differential routine in assembly.  :)
>=20
> Anyways hope this helps,
> Aaron Tufgar
>=20
> --
> http://www.piclist.com hint: To leave the PICList
> mailto:piclist-unsubscribe-request@mitvma.mit.edu
>=20
>=20
>=20

--
http://www.piclist.com hint: To leave the PICList
mailto:piclist-unsubscribe-request@mitvma.mit.edu