Noise and speaker adaptation techniques are essential to realize robust speech recognition in real noisy environments. We proposed that a noise robust speech recognition is implemented by superimposing a small quantity of noise data on spectral subtracted input speech. We also apply this noise robust speech recognition to the unsupervised speaker adaptation algorithm based on HMM sufficient statistics in different noise environments. According to spectral subtraction and noise superimposition, our proposed algorithm can make robust against the change of noises and SNR, and adapt quickly without calculating HMM sufficient statistics from noise matched acoustic models. We evaluate successfully our proposed algorithm with 20 k dictation task using four kinds of noises. The recognition experiments show that our proposed method increases the robustness against different noises significantly. We also compared our proposed method with unsupervised MLLR adaptation.
展开▼