band-pass filter for voices

[T o n g]

Hi,

Several noob-acoustic questions,

I'm trying to get from sox some aid to choose good parameters for band-
pass filters.

The http://img15.imageshack.us/img15/6717/spectrogram.png spectrogram
graph is produced with:

 sox rec_2009-04-24.wav -n spectrogram

I also try to use the '--plot' as:

 sox --plot gnuplot rec_2009-04-24.wav -n > rec_2009-0.gp

but it doesn't work. Why? How can I get a frequency spectrogram without
the time axis?

This will work,

 sox --plot gnuplot rec_2009-04-24.wav -n band 3k > rec_2009-0.band3k.gp

but how should I interpret the result graph
(http://img188.imageshack.us/img188/5667/screenshothd.png)?

The recording is a voice conversation in a small room. The problem is
that the echo from the walls make it very hard for me to make out what
was saying. I hope that the band-pass filter might make the voices crisp
again. but I don't know how to give good parameters for it, and how to
use sox's tools to find good parameters.

On trying to interpret my (posted) spectrogram myself, am I right that
the voices were within 2.2 to 2.8 KHz range, and 'bank 2.5k 0.3k' should
be used? Or, should I use sinc for a bandpass filter with steeper
shoulders?

Generally speaking, what's the good choices for band-pass filter to voice
recordings?

Thanks

--
Tong (remove underscore(s) to reply)
  http://xpt.sourceforge.net/techdocs/
  http://xpt.sourceforge.net/tools/


------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
Sox-users mailing list
Sox-users@...
https://lists.sourceforge.net/lists/listinfo/sox-users

[Fmiser]

> T o n g wrote:

> Several noob-acoustic questions,
>
> I'm trying to get from sox some aid to choose good parameters
> for band- pass filters.

> The recording is a voice conversation in a small room. The
> problem is that the echo from the walls make it very hard for
> me to make out what was saying. I hope that the band-pass
> filter might make the voices crisp again. but I don't know how
> to give good parameters for it, and how to use sox's tools to
> find good parameters.

You may be able to slightly improve the intelligibility with
audio filters - but it's too late to "solve" the problem. The
biggest issue is _not_ frequency spectrum related - but time
spectrum. The reflections from the room wall smear the arrival
of the consonants. And it is the consonants that are critical for
intelligibility.

In other words, the "mud" you hear is almost certainly "time
mud", not "frequency mud" and so a change in the frequency
response will only somewhat help.  So far as I know, there are
no effective tools for cleaning up time smear. :)

> On trying to interpret my (posted) spectrogram myself, am I
> right that the voices were within 2.2 to 2.8 KHz range, and
> 'bank 2.5k 0.3k' should be used? Or, should I use sinc for a
> bandpass filter with steeper shoulders?

Steep shoulders will probably not help any. Most of what we need
for intelligibility is between 1 kHz and 10 kHz. The vowel
sounds will be between about 150 Hz and 1 kHz.

I think what will help you most is a high-frequency shelving
filter. Set it to boost about 10 dB at 3 kHz to 5 kHz.  

"treble 10 3.5k"

Or conversely, a shelf cut at 200 Hz to 400 Hz.

"bass -10 300"

> Generally speaking, what's the good choices for band-pass
> filter to voice recordings?

There is very little useful vocal energy below 100 Hz (or for a
high pitched voice, as high as 250 Hz). At the other end of the
spectrum, there's not much beyond 10 kHz, thought some sounds
have harmonics that extend beyond 20 kHz. Usually, 8 kHz is high
enough for high-quality, intelligible speech.

For further "playing" with spectrum and time, check out baudline.
http://www.baudline.com/ 
(*nix only)

--     Philip

------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
Sox-users mailing list
Sox-users@...
https://lists.sourceforge.net/lists/listinfo/sox-users

[T o n g]

Thanks a lot Philip, for your comprehensive explanation.

On Wed, 07 Oct 2009 13:11:51 -0500, Fmiser wrote:

> The biggest issue is _not_
> frequency spectrum related - but time spectrum. . .


--
Tong (remove underscore(s) to reply)
  http://xpt.sourceforge.net/techdocs/
  http://xpt.sourceforge.net/tools/


------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
Sox-users mailing list
Sox-users@...
https://lists.sourceforge.net/lists/listinfo/sox-users