Research On Speech Recognition

Speech recognition is very promising in application. As an interdisciplinary field, it is also theoretically very valued. In this paper, the author adopts Dynamic Time Warping (DTW) model and Hidden Markov Model (HMM) respectively, to study the little vocabulary and isolated speech recognition. And probe into the problem of how to project a speech recognition system base on the hardware, also the problem of pitch estimation.Theoretical model is significative to help design the speech recognition system. Firstly, this paper analyzes the framework of speech recognition system in different processing levels, to illustrate how to choose the appropriate scheme in different task. Then discuss the fundamental methods of every part in speech recognition, use DTW realize isolate speech recognition.HMM has advantage of establishing model for time series, so we can use HMM to establish models for each word. Compared with every word's model, a word's speech can be recognized. In this paper, we simulate and complied some sub-programs on the platform of Windows. These sub-programs, which include pretreatment, endpoint detection, feature selection, HMM initial value setting use K-means clustering Algorithm, speech templates training, hidden markov modeling searching, realized the whole course of speech recognition. Mandarin digital speech recognition experimentation proves this scheme is feasible.At the end of this paper, presents speech recognition achieved on the hardware platform. Based on the characters of FPGA, give simplify method of short-time Autocorrelation function for the pitch estimation, and discuss one bit quantify apply in the speech signal frequency analyze.