The Design Of The On-board Speech Recognition System Based On ARM

Speech recognition technology is one of the linguistics and digital technology combined with technology, is also a very complicated engineering, but also a more frontier discipline. Of course, because of the voice system is complex, the interference of outside noise, speech signal nonstationarity, poor adaptability and Chinese own a series of problems, so as to make the speech a lot of difficulties and problems in the implementation.In this paper the algorithm of speech recognition system and onboard speech recognition system hardware and software environment to do detailed discussion, the main contents are:1. To present the algorithm of speech recognition system, need to discuss the speech recognition system of collection, pretreatment, endpoint detection, feature extraction and pattern matching parameter and so on several common algorithms, according to the requirement, this paper choose the suitable parameters and algorithm based speech recognition system is necessary.2. The traditional endpoint detection algorithm:short-term average zero crossing rate and short time average energy made a meticulous research. According to their existing problems, and puts forward some improvement on algorithm research. Of the two improved algorithms based on pure speech signal and high signal-to-noise ratio can be completed under the condition of endpoint detection. However, in a low SNR environment will fail. According to the characteristics of these two algorithms, the improvement and determine short-term combined weighted zero crossing rate and short range detection algorithm as the endpoint detection algorithm in this paper. Change after the endpoint detection algorithm in the clean speech signal and the low or high SNR environment also can complete endpoint detection, and thus improve the antinoise ability of the system and the recognition rate.3. Choose Dynamic Time neat (Dynamic Time Warping) as a pattern recognition algorithm of this paper, at the same Time, for the traditional DTW algorithm to improve the shortage of the existing problems, improve the overall path constraint and starting point of slack end point of the alignment. Results show that the improved DTW algorithm significantly improve the running speed of the system.4. In order to meet the requirements of vehicle, and real-time performance. This thesis with ARM11S3C6410for embedded processor, based on Linux embedded operating system, to Qt for embedded visual operating interface, establish a speech recognition system hardware environment include speech recognition module LD3320chip, audio51level controller ATMEGA128chip UDA1341TS chip, etc., as well as the Boot Loader, the Linux kernel.According to the results of experiment and data analysis of the results:the improved endpoint detection algorithm improves the running rate, on average, to increase the antinoise ability of the system and the recognition rate; The improved DTW algorithm in recognition rate under the premise of basically remain unchanged, the operation of the system rate increased by about20%; This thesis on-board speech recognition system based on pure speech signal recognition rate is about90%, in the big noise recognition rate can reach80%, also meet the real-time and practicality.