Research On An Embedded Speech Command Processing And Recognition Method

Speech recognition technology as a convenient man-machine interaction interface has been applied to the development of handheld devices, voice navigation, intelligent voice robots, industrial control and other fields. Embedded technology generation has accelerated the voice commands recognition technology from the laboratory to the market, making speech recognition technology has penetrated into all aspects of life, such as voice commands recognition technology in the intelligent home of the applied research. Thus, the research on embedded voice command processing and recognition methods has important practical significance and broad application prospects.In this paper, it has an embedded speech recognition technology as its study content, enhancing the effectiveness and robustness of voice command recognition system as its target, and study on the key issues, technology embedded in the front-end pre-processing and optimization identification methods of voice command recognition system.(1) Aiming at the low efficiency of independent component analysis techniques in speech de-noising, this paper improves the core iterative process of the method. The experimental results show that the modified method improves the convergence speed of iterations by about 30%.(2) Aiming at the difficulties of the poor universality and the precipitation of detection performance at low SNR in the field of speech endpoint detection, this paper gives a new endpoint detection method, with power spectral entropy and energy as its feature parameters. Simulation experiments show that the correct detecting rate of the new method has been improved 7% in stationary noise (white noise), and improved 5% in non-stationary noise (pink noise).(3) Study on wavelet neural network in recognition of voice commands, and gives an improved method to train network. The experimental results shows that the improved method makes full use of the sample information, reduce system training time, greatly improve the recognition rate.(4) Research on the Linux system in the embedded speech command recognition system, and achieve a voice command recognition system for the home environment, using relaxation starting and ending points of dynamic time warping as the identification method, and using cross-platform Qt GUI as the development tools.