Title Vowel recognition by gesture modeling of pseudo-formants

 

Vowels are generally characterized by the first three-formant frequencies. Each of them can be represented in the acoustic space (F1-F2 or F2-F3 planes) by a dot. The first three formants (F1, F2 and F3) are known to be necessary and in most instances sufficient for the representation of vowels. They denote the frequencies of the first three resonant modes of the vocal tract, and they permit intelligible synthesis of vowels.

However, due to the nature of the sounds, it is often impossible to well characterize the formant evolutions during speech, specifically during consonant production. It is why, MICA researchers founded that SSC parameters (Spectral Subband Centroid features) could replace formants and act as “pseudo-formant” even during consonant production.

The purpose of the recognition of vowels is to find a vector that characterizes the best the SSC parameter evolutions. This vector, which is a representation of the signal, should allow discrimination among all the vowels, because it could help to characterize the acoustical gesture realized during speech production. Indeed, some research shows that listeners adjust their understanding and use these dynamic information to recognize vowels transitions.

A new model, proposed by René Carré, can perform experiments on the vowel-vowel transitions. In fact, if the two vowels composing the transition are not physically affected, and even if they are placed outside of the vowel triangle, the sound can be perceived in the same manner as if they were living as the direction and the transition slope are maintained: which means that the gesture as sound is maintained.

The main objective of this subject is to realize the model propose by René Carré to find a way to characterize the acoustic gestures in the case of vowel-vowel transition.

 

Work description:

- Firstly, students need to understand the theoretic background involved in this subject, learn to use the program that can detect automatically the acoustic trajectory (written in Matlab).

- Re-coding the program in C/C++, Validate and evaluate the C/C++ program with male/female voice.

 

Student prerequisites

- Engineer student (final 5th year) or Master student from computer sciences field.

- Vietnamese or international student

   
Supervisors/contacts

Prof. Eric Castelli: This email address is being protected from spambots. You need JavaScript enabled to view it.

Dr. Nguyen Viet Son: This email address is being protected from spambots. You need JavaScript enabled to view it.