Research And Implementation Of Fast Recognition Algorithm Based On Embedded Isolated Words

In recent years, with the pace of life is accelerating gradually and the quality of human life is continuous improvement, personal computers and other electronic products are becoming more and more important in our daily life. This led to the demand of human for electronic products continue to improve and the intelligent requirements are also getting higher. The computer is becoming increasingly important in our usual work and study, while people urgently hope that we can communicate with the computer more convenient effectively and fast. The speech recognition technology achieved that let the machine understand human language. With the speech recognition technology is constantly mature. We can operate the computer through the speech order without the keyboard in the near future.This paper firstly gives a brief introduction to speech recognition system, including speech signal pretreatment, there are speech signal pre emphasis, frames and windows, speech signal endpoint detection; feature extraction of speech signal, the feature parameters mainly introduces the linear prediction cepstrum coefficient and Mel frequency cepstrum coefficient(MFCC); Speech recognition algorithm module: the module is mainly introduced the principle of vector quantization, dynamic time warping algorithm, hidden Markov model algorithm. The main work of this paper focuses on two aspects, one is to put forward a kind of new time-frequency characteristics—zero-spectrum, which is used as the feature parameters of speech recognition, also put forward a new speech recognition algorithm recognition model which has high recognition rate and simple called the bionic pattern recognition algorithm based on the hyper-sphere string; another is the MATLAB simulation of speech recognition system and the realization of hardware system, mainly including the simulation of the speech signal recording, the simulation of speech signal short-time energy and short-time average zero crossing rate, simulation of endpoint detection, feature extraction and isolated word recognition; We use Samsung STM32F103RBT6 microprocessor for isolated word speech recognition system in this paper, the hardware and software design are given in detail, and complete the design of Isolated word speech recognition system.Through the research and comparison of traditional voice recognition systems feature extraction algorithm,this paper proposes a new method of feature extraction and use the bionic pattern recognition algorithm to identify in the recognition stage. The simulation results on MATLAB show that the algorithm runs fast and has high recognition rate. On the other hand, the whole algorithm is used to realize the isolated word recognition on STM32. Speech recognition technology products will continue to improve our quality of life, but the high cost, which caused a great impact on the popularity of the product. This research has a certain application value, under the premise of ensure the speech recognition rate and recognition speed, to achieve the overall design of low power consumption, low cost and easy to carry. It has a very good application prospect in the field of speech recognition.