Improved spectrum subtraction method based on human ear masking effect and Bayesian estimation

Abstract

The invention discloses an improved spectrum subtraction method based on a human ear masking effect and Bayesian estimation. The improved spectrum subtraction method comprises the steps of: (1) adopting an improved minimum control value recursive averaging algorithm to obtain noise power spectrum estimation of an original noisy speech; (2) combining the obtained noise power spectrum estimation forperforming preliminary spectrum subtraction on a noisy speech signal; (3) performing Bayesian estimation based on weighted likelihood ratio distortion measurement on the signal after preliminary spectrum subtraction, and calculating the optimal estimated amplitude spectrum of the signal; (4) calculating a subtraction parameter of secondary spectrum subtraction by utilizing the human ear masking effect; (5) performing IMCRA noise estimation again before secondary spectrum subtraction, and carrying out secondary spectrum subtraction to obtain a final enhanced speech signal; (6) and performing inverse Fourier transform on the enhanced speech signal to obtain a final enhanced speech. The improved spectrum subtraction method better guarantees the intelligibility of the speech while improving the noise elimination capability of the algorithm, thereby improving the overall effect of speech enhancement.

Description

technical field [0001] The invention relates to an improved spectral subtraction method based on human ear masking effect and Bayesian estimation, belonging to the technical field of speech signal processing. Background technique [0002] Speech is an important way of information exchange between people, but people are always disturbed by various noises in the process of using voice to communicate and communicate. Noisy speech will not only increase human hearing fatigue and reduce the quality of speech communication, but also degrade the performance of the speech processing system based on feature parameter extraction. Therefore, in order to reduce the impact of background noise on speech quality, it is necessary to perform speech enhancement to suppress background noise. [0003] Spectral subtraction is a traditional enhancement algorithm. Its basic idea is to calculate the short-term amplitude spectrum of the noisy speech signal and the short-term amplitude spectrum of t...

AI Technical Summary

Problems solved by technology

[0006] Combining the auditory masking effect of the human ear is an important idea to eliminate the musical noise of spectral subtraction. Someone improved the spectral subtraction formula to: Y w (ω) is the spectral signal of noisy speech, In order to enhance the spectral signal of speech, most of the noise estimation algorithms used in existing technical solutions are not accurate enough, such as voice activity detection (VAD) or minimum value statistics, the reliability of the former will decrease with the decrease of signal-to-noise ratio, The latter responds slowly, which will affect the accuracy of noise estimation and reduce the effect of speech enhancement.

Moreover, there is also a misunderstanding in the current technical solutions. Too much emphasis on the elimination of music noise affects the intelligibility of the voice signal, destroys the voice signal, and even reduces the signal-to-noise ratio.

Images