Speech-independent Isolated-word Recognition Base On HMM

Speech recognition technology is a branch of speech signal processing, speech recognition technology turns the voice signal into text or commands with a machine which could identify and understand the voice signal. The theory of speech recognition has already very mature, and achieved a high recognition rate. Our speech recognition research started in the fifties, but has developed rapidly in recent years. The level of research has step from the laboratory into practical, however, as applied to the actual speech recognition, there are background noise, accent, speaking, and many other effects, that's the reason why this technology has not made a wide range of applications. And yet most of the products are still confined to laboratory environments, those research focuses on how to achieve online unsupervised learning and multiple methods integrated adaptive learning algorithm, and how to improve the recognition accuracy and reduced system complexity are the other purpose of those research.Since the research on speech recognition of the lab has just started, the foundation of large vocabulary speech recognition system need to create dictionary which requires a lot of linguistic knowledge and also need a big speech database, the paper mainly studies speaker-independent continuous Chinese digit strings speech recognition, including research on adaptive endpoint detection, contribution of Mel Frequency Cepstrum Coefficient(HMM)components to recognition rate, choice of numbers of HMM status and size of train set.Speaker-independent isolated-word recognition system based on hidden markov model(HMM) is presented. According to the calculation and analysis, we improve the methods of endpoint detection by setting double threshold and the real-time updates of threshold, The calculate time of system can be optimized by reducing the sampling rate, the number of states of each syllable, extending the frame to be recognized voice frame shift. The system real-time experiments show that the optimized speech recognition system is more able to accurately extract the desired speech segments, calculate time is dramatic declined, while the recognition rate decrease is minimal. The algorithm of the paper are realized by principle prototype based on PC104 embedded system, and implemented actual testing and validation in a laboratory environment, the results show this system has a high recognition accuracy, reliability, and real-time.