A method for enhancement of background sounds in forensic audio recordings

A method for suppressing speech while retaining background sound is presented in this paper. This procedure is useful for audio forensics investigations in which a strong foreground sound source or conversation obscures subtle background sounds or utterances that may be important to the investigation. The procedure uses a sinusoidal speech model to represent the strong foreground signal and then performs a synchronous subtraction to isolate the background sounds that are not well-modeled as part of the speech signal, thereby enhancing the audibility of the background material.