Adaptive acoustic masking based on spectral envelope

Acoustic masking can be used to protect the voice information from being eavesdropped. This paper presents a method to generate masking signal which is similar to white noise. It adjusts the white noise spectral energy by linear prediction of the masked speech spectral envelope in real time. The proposed method can accurately determine the frequency spectrum of the masking speech signal, and can adaptively adjust the masking noise volume by detecting masked speech energy in time domain. This algorithm was tested by using SII (Speech Intelligibility Index) defined in American National Standard ANSI S3.5. Experimental results show that this scheme provides better acoustic comfort and better masking capability compared with white masking noise. The algorithm can significantly improve the difficulty of stealing useful voice information by eavesdropper.