LINGUIST List 11.2324

Fri Oct 27 2000

Sum: LPC Resynthesis

Editor for this issue: Marie Klopfenstein <marielinguistlist.org>

Directory

1. peter dennis viechnicki, Summary: LPC resynthesis with altered formant frequencies

Dear Colleagues,

A few weeks ago I posted a query about LPC resynthesis with altered
formant frequencies. Here is the original query:

>I am searching for a way to re-synthesize speech from LPC parameters, but
>with altered formant frequencies. I know this has been accomplished
>before by certain researchers, for example Houde and Jordan in their paper
>on altered acoustic feedback, but I have been unable to locate any
>software or hardware that performs this function. Furthermore, I have
>been unable to find a straightforward way of modifying the lpc parameters
>by hand, in order to achieve the desired results with the formant
>frequencies. Can anyone provide advice or direction for this problem?


I received 5 helpful replies, from Chilin Shih, Hartmut Traunmuller,
Hugo Quene, Paul Boersma, and Joe Olive. Below are excerpts from
their replies.

Hugo Quene:

In response to your question posted to the LINGUIST List (11.2054) you may 
find the following information useful:

1. The comp.speech FAQ at
http://svr-www.eng.cam.ac.uk/comp.speech/Section3/Q3.2.html
gives an overview of links regarding speech coding, such as LPC.

2. One speech processing program, Praat, http://www.praat.org has the 
functionality of changing formant frequencies -- although not very 
user-friendly. If you hear about other software which does the job, I would 
appreciate to hear about it!

3. Without checking any further references, I think that you can find 
adequate recipes for changing formant frequencies in the 
male/female/gender/speech literature. I know about an article published by 
Klatt and Klatt in JASA, around 1990, and an article by I. Karlsson (1991) 
Female voices in speech synthesis, J. Phonetics 19, 111-120.


Hartmut Traunmuller:

In our lab at Stockholm University, we have had the possibility to do this 
since the 1970-ies, within the frame of certain restrictions, using the 
ILS-program package with certain additions by our own. After some 
extensions, I made use of the method in modifying the apparent age and sex 
of speakers for illustrative purposes (1989, see 
http://www.ling.su.se/staff/hartmut/manipul.htm ) and in an experimental 
study of the contribution of various acoustic factors to the perception of 
children's age and sex (1994). The method requires formant frequencies and 
bandwidths to be calculated from the reflection coefficients, and vice 
versa. It involves accepting substantial errors in bandwidths that are 
inherent in the LPC method, but our ears appear not to be very sensitive to 
this kind of distorsion. Anyway, I am now trying to develop a method in 
which these errors are reduced.


Joe Olive:

you can do it by getting the formants out of the reflection coefs and
changing them. If this is only done for a few samples it is quite
easy. you can solve for the roots of the lpc polynomial and some of
those roots are formants. You can change only the appropriate
ones. You may also want to look at an old paper of mine in JASA
Oct. 92.


Paul Boersma:

Paul Boersma writes that his software package PRAAT can indeed modify
formant frequencies for resynthesis. For more information on PRAAT,
which is available free of charge, please contact Boersma directly at:
paul.boersmahum.uva.nl.

Thanks to all who responded! Sincerely,
-Peter Viechnicki
University of Chicago Linguistics Department