[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Res: Re: synthesis/analysis: where is the error?


Thank you all very much! The responses are really helpful.


About the sounds I intend to use:


pitch: 115 Hz


F1 Continuum: from 300 Hz to 700 Hz in steps of 10 Hz.


Formant values: (300-700);  1160; 2500; 3500; 4500...(up to F10)

Formant Bandwidths: 50,100, 150, 200, 300, 400, 500, 600, 700, 800 (is it OK?).


Duration: 300 ms; rise/fall time:25 ms.


The program I am employing both for analysis and synthesis is Praat (for I dont have a Klatt synthesizer). The algorithm for the formant analysis is the Burg algorithm. The window is gaussian-like with sidelobes below -120 dB (whatever it means!) and lenght of 0.025 seconds.

The sampling frequency is 10050.

Some examples:

(F1 value specified in the synthesis/obtained F1 value in the analysis)











I have maked two vowel continua: the first one is a F1 continuum from 300 Hz to 700Hz (step size = 10 Hz) simply as synthesised by Praat. The second one is the first one adjusted according to the analysis results (assuming that the discrepancy between  synthesis and analysis results from some kind of interaction among the formants and between the formants and F0 and other parameters).


If the analysis is the main source of the discrepancy I can then use the first continuum assuming that the synthesis gives me the "right " F1 values, right?



-------Mensagem original-------
Data: 07/25/05 08:40:23
Assunto: Re: synthesis/analysis: where is the error?
Hello Daniel,
I would guess that the problem is in the analysis. The problem is that
if you use a fourier type analysis method the results can be way off
depending on the size of the data that you are using and if you are
using a non-rectangular window. I suggest that you modify the analysis
routine so that the size of the fourier window is an integer function of
the F1 in each case.
Let me know of one particular case.
1. What method are you using for the analysis.
2. What window are you using? Hamming, rectangular, cosine ??
3. What is F1, F2 etc. ?
4. For each freq in (3) give amplitude and phase
5. What is the length of your fourier window in samples ?
6. What is the time difference between samples ?
7. What is the result of your analysis ?
Daniel Silva wrote:
> Dear list,
> When I synthesize  a vowel contiuum (varying F1 in steps of, say, 10 Hz) and
> then analyse each sound I always see that there are differences between the
> F1 value specified in the synthesis  and the F1 value obtained in the
> analysis. Furthermore, ther is no linear relation between the two throughout
> the continuum.
> Where is the error? synthesis? analysis? bouth? Which is more reliable?
> If I want to make a continumm with a fixed step size...
> Should I rely either on the values I specify on the synthesis or on the
> observed values of the analysis?
> Thanks!
Fred Herzfeld, MIT '54
78 Glynn Marsh Drive #59
Brunswick, Ga.31525