SPoSS - Sound Patterns of Spontaneous Speech La Baume-les-Aix, France September 24-26, 1998

Multilingual Detection of Specific Sound Patterns: The Case of Vowels and Plosives

Nathalie Parlangeau, Francois Pellegrino

IRIT - Toulouse, France

This paper deals with the problem of multilingual automatic labelling. The proposed approach is based on unsupervised speech analysis and it aims to locate vowel segments and unvoiced plosive bursts. No learning phase is required and the algorithms are speaker and language independent. The vowel detection algorithm is based on a spectral analysis in Mel frequency filters and the plosive detection integrates a similar analysis with additional rules. Experiments are performed with Spanish and French data from the OGI Multilingual Telephone Speech corpus. Correct identification rates reach more than 90 % for vowel segments and near 85 % for plosives for both languages according to the semiautomatic labelling provided by OGI.

Full Paper

Bibliographic reference.  Parlangeau, Nathalie / Pellegrino, Francois (1998): "Multilingual detection of specific sound patterns: the case of vowels and plosives", In SPoSS-1998, 153-156.