Performance comparison of several speech recognition methods
Cuntai Guan, Ce Zhu, Yongbin Chen, Zhenya He · 2002
The paper compares several hybrid speech recognition methods an a highly confusable recognition vocabulary set. To characterize spectral variability of speech in speech recognition, Dynamic spectral warping (DSW) method is combined with DTW, HMM, and LVQ2 models. With these combinations, the discriminative abilities of these models are improved. Recognition accuracy is improved when these modified methods are tested on a speaker-dependent Chinese consonant recognition task. A TDNN model is also tested on the same database for comparison.>