Towards Speaker and Environmental Robustness in ASR: The HIWIRE Project

Alexandros Potamianos (1) , Ghazi Bouselmi (2) , Dimitrios Dimitriadis (3) , Dominique Fohr (2) , Roberto Gemello (4) , Irina Illina (2) , Franco Mana (4) , Petros Maragos (3) , M. Matassoni (5) , Vassilis Pitsikalis (3) , J. Ramirez (6) , E. Sanchez-Soto (1) , J. Segura (6) , P. Svaizer (5)

1 E.C.E - Department of Electronic and Computer Engineering [Crete]
2 PAROLE - Analysis, perception and recognition of speech
3 School of E.C.E - School of of Electrical and Computer Engineering [Athens]
4 LOQUENDO - LOQUENDO
5 ITC - Istituto Trentino di Cultura
6 UGR - Universidad de Granada = University of Granada

In this paper, we present algorithms for dealing with variability and mismatch in speech recognition due to environmental conditions and non-native speaker populations. The proposed algorithms cover a broad spectrum of ideas including robust feature extraction, feature compensation and speech enhancement. Specifically the following algorithms are presented and evaluated: beamforming for multi-microphone speech recognition, robust modulation and fractal features, Teager energy cepstrum coefficients, parametric feature equalization, speech enhancement, and acoustic modeling for non-native speech recognition. Also the problem of feature fusion and voice activity detection are discussed. Evaluation results on the AURORA databases under the auspices of the HIWIRE project show that significant gains can be achieved under adverse or mismatched conditions using these algorithms. Relative error rate reduction of up to 50% was shown for multi-microphone speech recognition, robust feature combination and speech enhancement. 30-40% reduction was shown for parametric feature equalization and non-native acoustic models.

Mots clés: speech recognition; noisy sppech; robustness

Domaines: Interface homme-machine [cs.HC]

Alexandros Potamianos, Ghazi Bouselmi, Dimitrios Dimitriadis, Dominique Fohr, Roberto Gemello, et al.. Towards Speaker and Environmental Robustness in ASR: The HIWIRE Project. SRIV'06 ITRW on Speech Recognition and Intrinsic Variation, May 2006, Toulouse, France. ⟨hal-00110502⟩