Speech recognition in the blind condition based on multiple directivity patterns using a microphone array

A novel hands free speech recognition method using a microphone array is proposed and is applied to the multi-talk recognition in the blind condition, no prior information about the sound sources and the characteristics of room acoustics. The proposed system is constructed by the cascade of the sound localization system, MUSIC, and the sound segregation system, SMDP (Segregation using Multiple Directivity Patterns) proposed in our previous paper. SMDP is characterized by using redundant directivity patterns. Usually, it is difficult for this sort of cascade system to achieve high performance because the sound localization stage cannot be perfect and errors occurred in this first stage cause serious damages to the segregation stage. Particularly missing the sound source is critical. By arranging the virtual sound sources, we treat the excess sound sources. In the proposed method, contrarily, the errors in the localization stage hardly cause the problems as long as they are insertion. SMDP uses redundant directivity patterns from the first, so it tolerates the insertion errors. The proposed method achieved 70% word accuracy in the double-talk recognition experiment of 20 K vocabulary, which is 18 point better compared to the ICA-based blind source separation with the source-number-given condition.