Title :
Noisy adaptive cepstral coefficients and its application to noisy speech recognition
Author :
Lee, Lee-Min ; Chen, Jen-Kwang ; Wang, Hsiao-Chuan
Author_Institution :
Dept. of Electr. Eng., Nat. Tsing Hua Univ., Hsinchu, Taiwan
Abstract :
The paper proposes a nonlinear cepstral equalization method for speech recognition. The method is based on AR modeling of clean speech power spectrum and the adding noise power to the speech. A noise ratio is introduced to provide a mechanism for adapting the reference template. An iterative algorithm is proposed to find a near optimal adaptation of reference cepstral parameters. Experiments showed that the proposed method is superior to the projection approach under severe noisy environment
Keywords :
acoustic noise; equalisers; hidden Markov models; linear predictive coding; optimisation; random noise; search problems; spectral analysis; speech coding; speech recognition; stochastic processes; time series; white noise; AR modeling; adaptation; adding noise power; clean speech power spectrum; iterative algorithm; near optimal adaptation; noise ratio; noisy adaptive cepstral coefficients; noisy speech recognition; nonlinear cepstral equalization method; projection approach; reference cepstral parameters; reference template; Cepstral analysis; Linear predictive coding; Noise level; Signal to noise ratio; Speech analysis; Speech enhancement; Speech recognition; Testing; Vectors; Working environment noise;
Conference_Titel :
Speech, Image Processing and Neural Networks, 1994. Proceedings, ISSIPNN '94., 1994 International Symposium on
Print_ISBN :
0-7803-1865-X
DOI :
10.1109/SIPNN.1994.344895
