- Registration time
- Last login
- Online time
- 31 Hour
- Reading permission
The paper mainly studies on the technology of speech command recognition based on aerocraft in the aerial domain. The purpose is to enrich the control mode of aerocraft, to lighten the pilot handling intensity, to improve manipulative speed and security. The paper has important practical significance and value to development of our country aviation project.
First, the paper analyzes the reasons that the detect performance of the traditional speech endpoint detection algorithm to apply in the aerial background dropped, studies the aerial noise in the intensity and type of noise, and gives a word boundary detection algorithm for variable noise environment The simulation result indicates that this performance of the endpoint decetion algorithm surpasses the traditional algorithm obviously under every kinds of noise environment. Moreover, the paper studies the traditional feature extract algorithm of MFCC in the computational complexity, finds that its computation is very big, which affects system real time. For the problem the paper improves the algorithm, ascertain the algorithmic optimum parameter by many experiment. The improved algorithm reduces the computation to half, corresponding recognition rate only cuts down 1.5%, which disregardes the effect to applying need.
The paper also introduces the hidden Markov model technology in speech recognition's application, studies the algorithmic basal principle and realization method of the acoustics model, language model and searching algorithm. For traditional Viterbi beam searching algorithm using fixed pruning threshold value's shortcoming, the paper uses the self-adjusting pruning threshold searching algorithm.
Finally, the paper constructs one small continuous speech command recognition system based on aerocraft with Visual C++, confirms the optimization parameter. This speech command recognition system's average recognition rate achieves 98.5%.
Key words: continuous speech recognition, speech command, endpoint detection, MFCC, aerial noise, search algorithm