[music-dsp] Is 10-bit ADC input good enough for speech?
joshscholar at yahoo.com
Tue Jan 13 04:52:00 EST 2004
By the way, my experience with 8 bit sound is that dither noise helps a lot.
Adding some blue noise before quantizing can improve your signal a lot when
you don't have as many bits as you might wish.
The result is greater dynamic range in everything lower than high treble at
the expense of having some high frequncy noise.
Audiophile quality samplers do this for you, but I doubt your10 bit sampler
has dithering built in.
As a test of algorithmic dithering (converting between formats), I once
wrote a test where I used a blue-noise error dispersion dither to play a
song in 1 bit sampling, with the peak volume 24 DB below the noise floor.
With a 44100 sampling rate the song was understandable, though I probably
injured my ears with the loud high treble noise. There was a very loud
22khz component that I couldn't even hear, but my ears rang afterwords.
----- Original Message -----
From: "Yaakov Stein" <yaakov_s at rad.com>
To: <music-dsp at shoko.calarts.edu>
Sent: Monday, January 12, 2004 11:03 PM
Subject: RE: [music-dsp] Is 10-bit ADC input good enough for speech?
> > > So: is 10-bits enough to record the human voice "cleanly"?
> For 8000 samples per second speech, 12+ bits is considered
> to be "toll quality" (i.e. score of >4 on a scale of 1 to 5).
> (Of course these can be squeezed into 8 bits by companding.)
> Using fewer than 12 bits adds perceivable background noise,
> but it may be good enough for some applications.
> Why don't you take a few clean wave files at the sampling rate
> you intend using, and either quantize them to 10 bits or add the
> amount of white noise? If you are happy with what you hear,
> then go for it.
> Jonathan (Y) Stein
> dupswapdrop -- the music-dsp mailing list and website:
> subscription info, FAQ, source code archive, list archive, book reviews,
More information about the music-dsp