Noise Robust Speech Recognition based on Spectral Reduction Measure
It has been reported that spectral features of spontaneous speech shrink from those of read speech and that this spectral reduction causes degradation in speech recognition performance. This spectral reduction is expected to be observed with other factors than spontaneity, and cause degradation. In this study, we analyze spectral reduction in noisy speech and its correlation to speech recognition accuracies. Based on the analysis result, we propose a model adaptation method to noisy environment. In our evaluation using noisy speech data, our method increased the recognition accuracies by 2.6 points.
Mayumi Beppu Koichi Shinoda Sadaoki Furui
Tokyo Institute of Technology, Tokyo, Japan
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-4
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）