Speaking rate estimation for multi-speakers

It is important to estimate speaking rates of multispeakers in multi-participants conversational speech, especially speaking rate of dominant participant. This paper proposes an algorithm for estimating speaking rates of multi-speakers. In the proposed algorithm, speaker segmentation and clustering are first performed. As a result, number of speakers and the corresponding speech of each speaker are obtained. Finally, detecting the local maxima of energy envelope of each speaker´s speech and then speaking rate of each speaker is defined as total number of local maxima divided by length of each speaker´s speech. Experimental results show that the proposed algorithm can estimate speaking rates of multi-speakers with satisfactory results, whereas the previous algorithms can only estimate speaking rate of single speaker.