Voice enhancement method based on continuous noise estimation

摘要: The invention discloses a voice enhancement method based on continuous noise estimation. In logarithmic spectrum domain, model which is trained in advance used to carry out estimation parameter of background noise, and an estimated mean value recover clean voice. Firstly, acoustic pretreatment fast Fourier transform are performed input so as acquire amplitude phase position each frame digital voice; the for subtraction recovering time domain signal. And then, sub-band filtering logarithm taken operate spectrum, extract from characteristic vector containing real time. Finally, weighted inverse overlap add enhanced invention, according frames changes tracked