The Application of Speech Recognition System in Noise Environment

The difference of training circumstance and recognition circumstance will mismatch the mode and test data, so normal speech recognition system doesn't work well in practical situation. There are many reasons of mismatching testing and training in speech recognition. Only the influence of interchannel noise and background noise is involved in the paper. The speech recognition in the case can be called de-noised speech recognition. In allusion to speech recognition of the signal from the sink of frequency-modulated station, the paper presents the speech recognition system in noise environment. The first grade of the system is two-stage Wiener filtering system, the post-grade speech recognition algorithm adopts way of DTW template matching, the library of mode is composed of pure speech of lab. After the non-real-time simulation test, it can be declared that: first, there is no obvious difference between the speech without been enhanced through recognizer and the speech enhanced through recognizer. And it shows that the pre-speech enhancement system doesn't effects the function of speech recognition system when the signal-to-noise performance is high; second, when the signal-to-noise performance is low, the correct rate of concatenation speech recognition system in noise improves obviously. What satisfactory the speech recognition system in noise.