Speaking rate control based on time-scale modification and its effects on the performance of speech recognition

In this paper, we describe the influence of speaking rate on speech recognition. Speaking rate of input speech is controlled by applying a time-scale modification (TSM) algorithm and speaking rate normalisation is achieved by selecting a scale factor of TSM. The scale factor selection for training and testing of a speech recognition system is performed based on a maximum likelihood criterion during HMM decoding. From the experimental results, we showed that optimal selection of a TSM scale factor in speaking rate normalisation can reduce WER by 47.6% compared to the baseline.