Research And Implementation Of Speaker-independent And Isolated Words Phonetic Recognition Based On DDP Platform

Speech interactive is the most important part of human-computer interaction. The most key technology is voice recognition to achieve the goal. However, due to the diversity of speech signal and the complexity of the speech recognition process, the speech recognition research progress both in theory or in the productization are difficult. There are still many challenging task. Relevant speech recognition products are hard to attain perfect.Today, Mobile Internet tide has flocked to. And speech recognition related products still mostly based on the desktop PC. And Apple Inc. just launched a new speech recognition product called Siri, as the main function of the cell phone iPhone4S recent release. It would be a rash that speech recognition products based on mobile equipment platform in the near future. This subject is the research and analysis and optimization for speech recognition based on mobile devices processor, and implement a small scale of speech recognition system.This subject adopts the HMM algorithm. The algorithm considers the probability of voice and audio source corresponding relation and audio source of state transition probability between elements, simulate speech recognition of the human with a double random process, achieve large vocabulary, speaker-independent, continuous speech recognition. The algorithm put forward in the80s, Dr. Kaifu Li from Carnegie Mellon university first realized the system based on the algorithm later. After that, the speech recognition algorithm has not from HMM framework.HTK is HMM Toolkit. It consists of a set of library modules and tools available in C source form. It is used for building and manipulating hidden Markov models. HTK was developed by Cambridge University Engineering Department at first, and then was owned by Entropic Company. In1999Microsoft bought Entropic Company and devoted HTK. to Open Source Community. Microsoft therefore decided to make the core HTK toolkit available again and licensed the software back to CUED so that it could distribute and develop the software.The algorithm of speech recognition is complex, which needs lots of floating-point calculation and higher performance requirement, thus the ordinary mobile processor is not qualified. Then, DSP processor, higher performance and especially suitable for floating-point calculation, is the most qualified choice. This subject adopts the high performance DSP processor of floating point named TMS320C6722of the TI Company, and match with the external equipment such as audio chips, expanded memory and so on, and realizes a speech recognition system of independent person with finite words. Finally analyzes the recognition results and puts forward the direction.