Design of Isolated Word Recognition Based on Hidden Markov Model

• Main Authors: Sheng-Wen Liao, 廖陞汶
• Other Authors: Yen-Tai Lai
• Format: Others
• Language: en_US
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/75841343046835186007

碩士 === 國立成功大學 === 電機工程學系 === 87 === The left-to-right continuous hidden Markov model (CHMM) with kernel using the viterbi algorithm has successfully been used for speech recognition in the past years. The viterbi algorithm includes two main tasks. One is the Gaussian calculation and the other is the optimal estimation of state path. However, general processor can not efficiently process the variance of arithmetic operations. We must design a VLSI hardware architecture of high speed to overcome the problem. In this thesis, we use the data scheduling method to create the architecture implementing the viterbi algorithm. The design is divided into Gaussian module and viterbi-kernel module to reduce the complexity of design. We implement the system by using standard cells technology. Experimental results show our method is effective.