Re: Sound Analysis Tools

[James Beauchamp <jwb@xxxxxxxxxxxxxxxxxxxxx>]

On Fri Nov 17 10:11:35 1995) Thierry Rochebois wrote:

>What I have is a frequency sampled gaussian.
>On a log scale a gaussian is a parabola.
>So, the quadratic interpolation of this spectrum on a log scale gives
>an accurate value of the frequency and amplitude.

We are using a Kaiser window. Isn't a Kaiser close to a guassian? After all
true guassians would require infinite time and frequency windows.

>For each frame I store the frequency, phase and amplitude for each peak
>of the spectrum (ie for every sine wave).
>Before resynthesis, I select the peaks that correspond to harmonics
>(thanks to an harmony criterion).

We've had good results using this approach for harmonic sounds. We have a pitch
detector program that makes this possible for variable fundamentals within a
limited search range. However, the program sometimes makes mistakes. Also, one
must decide on a threshold for the tracking method, and this can cause
problems.

For sounds of fixed pitch, I have had better and more consistent
results using a pitch-synchronous phase vocoder program.

Also, more recently I've been interested in tracking inharmonic partials of
percussion sounds, which the McAulay-Quatieri method can accomplish as long
as the partials are not too close together. Chime tones work well for example,
except the resynthesized attacks are a bit mushy. Cymbals and drums are another
story. I haven't figured out a way to stretch them using this approach without
introducing artifacts. I feel that the key is to latch onto the correct modal
frequencies.

Again on Fri, 17 Nov 1995 19:08:51 +0100) Thierry Rochebois wrote:

>After extracting the harmonic part from the signal, I can study the signal
>with other time frequency/resolution tradeoff. This can be efficient to study
>transients (high time resolution, low frequency resolution).

This is important. Two comments: 1) Most musical signals can tolerate a fair
amount of phase distortion if it is fixed. Witness what happens when we play
back in a room. Room modes cause all kinds of phase distortion. But preserving
the correct phases is very important for subtraction in the time domain. 2) We
shouldn't restrict ourselves to harmonics. Any well-defined partials could be
in the primary (non-residue) part of the signal.

Incidentally, the software I use for analysis, graphics, modification, and
synthesis on Unix workstations is available for FTP. Anyone who is interested
in a copy can contact me, and I will give you the acquisition information.

Jim

James Beauchamp
professor of music and electrical & computer engineering
2136 Music Bldg. MC-056, University of Illinois at Urbana-Champaign,
1114 W. Nevada, Urbana, IL 61801
email: j-beauch@uiuc.edu
phone: +1-217-244-1207/344-3307
fax: +1-217-244-4585
WWW:  http://cmp-rs.music.uiuc.edu/people/beauchamp/