Research And Implementation Of Embedded Speech Recognition

With the development of technology, current speech recognition technology under most laboratory conditions has high recognition rate. How to make speech recognition more mature for business applications and serve people is the needs of the times. Therefore, the research on this is imperative. If the voice recognition can be applied to embedded development, greater economic benefits will be brought.However, the implementation of speech recognition on embedded systems is also subject to many factors. For example, the computing environment for embedded systems is low in computing speed and limited in storage space. So the real-time requirement is difficult to meet. Although the speech recognition achieves some results on the high-performance platform, further analysis and research is still needed for application on embedded systems. This paper's main research works are listed as follows:Construct speech recognition hardware system, selecting high performance processor (DM6446) Dual-core processor as the key processing unit, and designing outlying work module base on it.Construct speech recognition software system with the U-Boot, file system and embedded Linux system. The research on speech recognition includes signal preprocessing, feature extraction, model library establishing, etc. And then the hidden Markov model is established to achieve the function of speech recognition. Based on the current hardware platform, voice recognition systems is tested, after analyzing the results, some conclusions are reached, that the recognition rate of frequently used words is higher than those less frequently used and homophones, and all of them could be improved after training.