DocumentCode :
2323960
Title :
Ear-model derived features for automatic speech recognition
Author :
de Mori, Renato ; Albesano, Dario ; Gemello, Roberto ; Mana, Franco
Author_Institution :
LIA CERI-IUP, Univ. of Avignon, France
Volume :
3
fYear :
2000
fDate :
2000
Firstpage :
1603
Abstract :
The paper provides a theoretical justification that gravity centers (GC) in frequency bands computed from zero-crossing information are far more robust to additive telephone noise than GCs computed from FFT spectra. Experiments on two different corpora confirm the theoretical results when GCs are added to standard mel frequency-scaled cepstral coefficients (MFCC) and their time derivatives. A 20.1% word error reduction is observed on a large telephone corpus of Italian cities, with an average signal-to-noise ratio (SNR) of 15 dB, if GCs are computed from zero-crossings, while performance deteriorates when GCs are computed from FFT spectra
Keywords :
acoustic noise; cepstral analysis; speech recognition; FFT spectra; Italian cities; SNR; additive telephone noise; automatic speech recognition; average signal-to-noise ratio; ear-model derived features; frequency bands; gravity centers; large telephone corpus; performance; standard mel frequency-scaled cepstral coefficients; time derivatives; word error reduction; zero-crossing information; Additive noise; Automatic speech recognition; Cepstral analysis; Frequency; Gravity; Hidden Markov models; Neural networks; Noise robustness; Telecommunication computing; Telephony;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Acoustics, Speech, and Signal Processing, 2000. ICASSP '00. Proceedings. 2000 IEEE International Conference on
Conference_Location :
Istanbul
ISSN :
1520-6149
Print_ISBN :
0-7803-6293-4
Type :
conf
DOI :
10.1109/ICASSP.2000.862002
Filename :
862002
Link To Document :
بازگشت