Temporal patterns (TRAPs) in ASR of noisy speech

Author

Hermansky, Hynek ; Sharma, Sangita

Author_Institution

Graduate Inst. of Sci. & Technol., Portland, OR, USA

Volume

1

fYear

1999

fDate

15-19 Mar 1999

Firstpage

289

Abstract

We study a new approach to processing temporal information for automatic speech recognition (ASR). Specifically, we study the use of rather long-time temporal patterns (TRAPs) of spectral energies in place of the conventional spectral patterns for ASR. The proposed neural TRAPs are found to yield significant amount of complementary information to that of the conventional spectral feature based ASR system. A combination of these two ASR systems is shown to result in improved robustness to several types of additive and convolutive environmental degradations

Keywords

feature extraction; feedforward neural nets; multilayer perceptrons; noise; pattern classification; spectral analysis; speech processing; speech recognition; additive environmental degradation; automatic speech recognition; complementary information; convolutive environmental degradation; feed-forward multilayer perceptron; long-time temporal patterns; neural TRAP; neural temporal pattern classifier; noisy speech; spectral energies; spectral feature based ASR system; spectral patterns; temporal information processing; temporal patterns; Automatic speech recognition; Communication channels; Computer science; Data mining; Databases; Degradation; Frequency; Robustness; Telephony; Working environment noise;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1999. Proceedings., 1999 IEEE International Conference on

Conference_Location

Phoenix, AZ

ISSN

1520-6149

Print_ISBN

0-7803-5041-3

Type

conf

DOI

10.1109/ICASSP.1999.758119

Filename

758119