EMD decomposition to discriminate nasal vs. oral vowels in French

Mohamed Réda Saïdi (1, 2) , Olivier Pietquin (2) , Régine André-Obrecht (1, 3)

1 IRIT-SAMoVA - Équipe Structuration, Analyse et MOdélisation de documents Vidéo et Audio
2 SUPELEC-Campus Metz
3 UT3 - Université Toulouse III - Paul Sabatier

In this work, we introduce a new parametrization, implying the Empirical Mode Decomposition (EMD), to discriminate nasal vs. oral vowels. The proposed method consists of three classical stages, signal preprocessing, feature extraction and decision. Firstly, the speech signal is decomposed using the EMD method which has the advantages of being adaptive and signal-length independent. Then, a Linear Prediction analysis is applied to some EMD components to provide the observation parameters (Linear Prediction Cepstral Coefficients (LPCC)). Finally, Artificial Neural Network (ANN), K-Nearest-Neighborhood (KNN) and Gaussian-Mixture-Model (GMM) classifiers were used to distinguish nasal vowels from oral vowels in French (French database Bref80). Over all decision methods, tests show an improvement when using the new parametrization compared with a standard LPCC analysis.

Domains: Computer Science [cs]; Numerical Analysis [cs.NA]

Mohamed Réda Saïdi, Olivier Pietquin, Régine André-Obrecht. EMD decomposition to discriminate nasal vs. oral vowels in French. SPECOM 2009, Jun 2009, Saint-Pétersbourg, Russia. (5 p.). ⟨hal-00437066⟩