The Android Platform Research And Implementation Of Isolated Word Speech Recognition Algorithm

Speech recognition is the technology which converts speech signal understanding and identification into the appropriate commands or text through the machine and allows the machine to perform text or instruction to complete human-computer interaction. The research object of speech recognition technology is speech signal, which involves many disciplines,such as psychology, linguistics, computer science and signal processing, etc. and is also important branch of pattern recognition. Meanwhile, under the historical background of booming development of big data in the mobile Internet, more applications of speech recognition technology based on intelligent terminals are emerging, which play an important role in various fields, such as the industry, household appliances, telecommunications,automotive electronics, medical treatment, family services and consumer electronics, etc.,therefore, it is of very realistic significance in the study of intelligent terminals speech recognition technology and the improvement of accuracy and timeliness of speech recognition.The author in this paper studies the speech recognition key technology, which mainly includes the following four aspects:1. Design and implement the speech information collection system based on Android platform. Realize the speech information collection, storage, playback of voice data and the whole software function module design on the Android platform, which lays foundation for further analysis of speech signal in the late period.2. Research the methods and techniques of speech signal preprocessing. Achieve the pre-filtering, pre-emphasis, framing windowing and endpoint detection of the speech signal.Filter the power supply noise in the speech signal and ambient noise; obtain the short-term and stable speech frame, thus calculating the speech signal which meets the extraction conditions of characteristic parameters.3. Research and improve the extraction algorithm of speech signal characteristic parameters. Focus on studying the extraction algorithm of characteristic parameters of Mel-frequency Ceptral Coefficients(MFCC) based on human auditory model; use half amplitude spectrum to the power spectrum, use the frequency resolution to initialize Mel triangular filter bank. Use Mel triangular filter response to do logarithmic cubic compression algorithm to achieve the algorithm improvement and enhance the performance of characteristic extraction algorithm to some extent.4. Research the speech recognition models. Focus on achieving two recognition algorithms, the dynamic time warping(DTW) and Hidden Markov(HMM). Conduct thetraining and recognition of Chinese speech characteristic parameters of number from 0 to9 through these two algorithms and use many people samplings to conduct experiments and then find that the recognition rate of Hidden Markov(HMM) is better than the dynamic time warping algorithm. The experimental result shows that the recognition rate of isolated word speech recognition system based on the algorithm proposed in this paper is about 60%, which basically meets the requirements of the system design.