A study of robustness in noisy speech recognition using weighted variance expansion of word HMMs

摘要

The spectra of noise and SNR often vary abruptly due to the nonstationary noise under field conditions. The performance of speech recognition degrades rapidly when the noise conditions in the recognition process are different from those in the process of training or adaptation, therefore it is necessary to make HMM robust to abrupt variation of noise. In this paper, we propose a method to modify the output probability in the states sensitive to noise by using weighted variance expansion based on the power of states or probability distributions, in order to improve the performance. The effectiveness of this method was confirmed in not only clean speech HMMs but also noisy speech HMMs, through the evaluation experiments of speaker independent word recognition using noise of 2 factories.