Research On Design Of Speech Recognition Soft IP For Isolated Words

Speech Recognition application is widely used in recent years due to rapid technology developments of digital signal processing and integrated circuit design and manufacture Speech Recognition System-on-Chip also appears. This thesis focuses on the soft IP design of speech recognition for isolated words.Vector Quantization algorithm is used in the design process, because of its recognition speed is more quickly. It can do real-time recognition although codebook number is very large. Furthermore, Vector Quantization algorithm need smaller storage to store codebook than other algorithms. This is it's another advantage. This thesis discusses the recognition effect and recognition speed, researches the connection between the algorithm and hardware realization. Although the design becomes easier and the circuit scale is smaller, the result also perfect.Finite State Machine is the main design method in the hardware design .To reduce the circuit size, two main methods are used. One is to adopt Schur algorithm at calculating LPC parameters. The other is to improve random sample initial codebook algorithm to replace Kmeans algorithm.In this dissertation, those tasks have been made. First, IP-based SOC design methodology has been discussed. Then the basic theory of speed recognition for isolated words and vector quantization algorithm are researched. Lastly hardware realization and its verification is experimented, which give a good result.