Research On FPGA-based Voice Enhancement Technology

With the rapid development of intelligent application fields,people try to control smart devices to complete complex tasks through voice commands.However,during the process of propagation and reception,speech is subject to different levels of noise interference,affecting the transmission and recognition of speech.Therefore,in order to reduce the influence of noise,enhance the voice signal-to-noise ratio,and improve the voice quality,it is necessary to perform noise reduction processing on the voice signal.Aiming at the problem that the basic spectral subtraction speech enhancement technology is not accurate for noise estimation,the median filtering and noise endpoint detection algorithms are introduced to improve the ratio of the average zero-crossing rate to the short-term energy value by using the time-frequency domain difference of speech and noise.Perform noise endpoint detection.Using FPGA parallel processing capability and design flexibility,the hardware design of improved spectral subtraction and LMS adaptive speech enhancement algorithm on FPGA is realized.A hardware test circuit is built for two voice enhancement technologies.Tests show that in the noisy speech signal model,the signal-to-noise ratio difference before and after improved spectral subtraction is about 4.1 dB for 10 dB stationary white noise,and the LMS adaptive speech enhancement algorithm is about 7.6 dB.The nonstationary noise test results show that the improved spectral subtraction method is suitable for a wider range,and the LMS adaptive algorithm is suitable for low SNR environments.In the actual environmental test under white noise interference,the improved spectral subtraction has better processing results than the LMS adaptive algorithm.