In order to overcome the shortcomings of traditional BP (Back Propagation) in the process of evaluation for continuous speech jamming effect, a method based on QGA (Quantum Genetic Algorithm) is proposed to optimize the BP network for objective evaluation of speech jamming effect. The MFCC feature of continuous speech data under different distortion conditions are extracted by the Mel cepstrum, which are set to the model inputs in the QGA-BP method. Then the QGA is used to optimize the weight and threshold of BP neural network from multiple spots. A QGA-BP model is constructed to accomplish the mapping from the input to the Mean Opinion Score (MOS). Representative speech data is used to verify this method, and the results show that the convergence speed of QGA-BP model is improved compared with the traditional BP network. Meanwhile, the convergence error can be effectively controlled within 4%, and the model correlation coefficient and the standard deviation are closer to the ideal parameters.

backpropagation; genetic algorithms; jamming; neural nets; speech processing; continuous speech jamming effect; continuous speech data; QGA-BP method; BP neural network; QGA-BP model; representative speech data; QGA-BP network; MOS; mean opinion score; distortion conditions; quantum genetic algorithm; Sociology; Statistics; Jamming; Genetic algorithms; Feature extraction; Distortion measurement; Cepstrum; Jamming evaluation; BP neural network; QGA; Mel cepstrum; Characteristic parameter

--