New method for detecting endpoint of speech in noise environment
Qingning Zeng · Shengxue jishu · 2007
By studying noisy speech in the time-frequency domain, a new speech detection method based on decomposition with matching pursuits is proposed. With matching pursuits, the decomposed time-freq-uency energy of the signal has a clear distribution in the time-frequency domain. Using the speech characteristics, endpoints can be differentiated by setting a threshold value. Compared with the conventional method that uses zero crossing ratio to detect the surd and uses short-time energy to detect the sonant, although the real-time capability is sacrificed, the accuracy of endpoint is greatly improved in noisy envir-onments.