Research on ASIC for multi-speaker isolated word recognition

The ASIC for multi-speaker speech recognition is design in this paper. The LPC-derived cepstral coefficients are chosen as speech features. Templates are trained by K-means clustering algorithm. Two stage recognition system can not only improve recognition accuracy, but also reduce the delay. The first stage of recognition system uses speech spectrum difference (SSD) algorithm. The second stage uses DTW. The whole recognition system is design into ASIC on high level with VHDL and simulated in Powerview.