Study Of Specific Vocabulary Speech Recognition System Based On OR1200

Speech recognition has a wide application prospect in intelligent home furnishing, human-computer interaction, military application, industrial control and other fields. Speech recognition system consists of endpoint detection, feature extraction and pattern matching.Feature extraction is the central part.The feature extraction algorithm mainstream of Linear prediction cepstrum coefficient and Mel frequency cepstral coefficients, that have relatively high algorithms, is basically based on the PC engine implemented in software form; Speech recognition system at home and abroad based on hardware implementation is mainly based on the embedded system.And the system that realizes the recognition of medium or small vocabulary, can be embedded into other systems as the communication interface between man and machine to achieve a simple interactive between voice and hardware systems. And the study of speech recognition chip used in intelligent home furnishing, industrial control and other fields, has a broad application prospect and significance.Speech recognition systems based on hardware implementation, have mainly 3 kinds of schemes:1. Build the hardware system based on ASIC; 2. Achieve a combination of hardware and software based SOC; the 3 is based on the hardware system, which has great flexibility.However. The SOC scheme in the current speech recognition technology is still not mature enough that can be used by the majority of people.This design adopts the method of the combination of hardware and software to design the speech recognition chip, based on the OR1200 processor. The feature extraction part is realized by hardware.However, endpoint detection and pattern matching are implemented by OR1200 software. Speech recognition algorithm is mainly composed of endpoint detection, feature extraction and pattern matching. The feature extraction part adopts the Mel frequency cepstrum extraction scheme, in which the time-frequency conversion adopts 128 points FFT algorithm.128 FFT with 2*8*8 mixed-radix structure is easy to handle with pipelining, and has good data throughput;While endpoint detection has a small amount of computation, pattern matching module has complex calculation and irregular structure.And both of them adopt the software approach implementation. The speech recognition module is connected with the OR1200 processor through the Wishbone bus, that the core of SOC is composed of. Through simulation, we can complete the the speech recognition of a few specific vocabulary.The paper studies the use of 32bit open-source processor OR1200 as the core processor to mount modules with Wishbone bus to design SOC chip.The design method of the chip takes into account of application flexibility, system speed and development cost, that is good for application and promotion. However, the system is applied in the intelligent home furnishing and other fields, having high requirements for the chip cost.Researching the whole system has still a long way to go.Our study provides a foundation for the research of the whole system, and a platform for future research.Meanwhile,our research provides a reference scheme for the future design of SOC chip.