Simultaneous recognition of distant talking speech of multiple sound sources based on 3-D N-best search algorithm
Tóm tắt
This paper deals with the simultaneous recognition of distant-talking speech of multiple talkers using the 3D N-best search algorithm. We describe the basic idea of the 3D N-best search and we address two additional techniques implemented into the baseline system. Namely, a path distance-based clustering and a likelihood normalization technique appeared to be necessary in order to build an efficient system for our purpose. In previous works we introduced the results of experiments carried out on simulated data. In this paper we introduce the results of the experiments carried out using reverberated data. The reverberated data are those simulated by the image method and recorded in a real room. The image method was used to find out the accuracy-reverberation time relationship, and the real data was used to evaluate the real performance of our algorithm. The obtained Top 3 results of the simultaneous word accuracy was 73.02% under 162 ms reverberation time and using the image method.
Từ khóa
#Speech recognition #Hidden Markov models #Viterbi algorithm #Search methods #Natural languages #Clustering algorithms #Reverberation #Feature extraction #Adaptive systems #SortingTài liệu tham khảo
heracleous, 1999, Simultaneous Recognition of Multiple Sound Sources based on 3 - D N-best Search, Proc of Acoustical Society of Japan, 91
10.1109/ICASSP.1998.674413
10.1121/1.382599
heracleous, 2001, Multiple Sound Sources Recognition by a Microphone Array-based 3-D N-best Search with Likelihood Normalization, Proc International Workshop on Hands-Free Speech Communication, 103
10.1016/0167-6393(95)00011-C
heracleous, 2000, A technique for likelihood normalization in the 3-D N-best search for simultaneous recognition of multiple sound sources, Proc of Acoustical Society of Japan, 117
10.1109/ICSLP.1996.607855
10.1109/ICASSP.1996.543272