Internet Explorer® recommended
Unvoiced Fricatives Synthesis
Some unvoiced fricatives were selected to be synthesized using MATLAB®. These phonemes were: /T/, /f/, /S/, /s/, which correspond to the English phonemes dictionary. Also, some sample words were recorded, in order to obtain real representations of the phonemes to be sintesized.
The analysis is based on frequency domain by taking the magnitude spectrum of the real phonemes as target shapes. The way to approach to the desired spectrums was using a bank of 2nd order resonant filters, which are in charge of shaping a white noise source in order to obtain the desired results
Besides, hybrids versions of the recorded words were made replacing the real phonemes by the syntheslized ones in order to generate an appropriate context to compare. If you want to go over the code, you can find it here.
The following table shows the number of formants, the respective bandwidths, the center frequencies for each resonator and links to the respective plot pages. However, the most interesting resource is in the Audiofiles column. You can play the "hybrid" and the real recorded words for each phoneme. Could you recognize which one corresponds to the synthetized version? (The answers are here.)
|
Phoneme |
Number of main formants |
|
Frequencies of the formants (Hz) |
Audiofiles (Comparison) |
Plots |
|
/f/ |
3 |
100,300 |
780 |
|
|
|
/T/ |
4 |
300 |
1988 |
|
|
|
/s/ |
3 |
100 |
1734 |
|
|
|
/S/ |
3 |
100 |
2605 |
|
