Self-Produced Speech Enhancement and Suppression Method using Air- and Body-Conductive Microphones

This paper presents a self-produced speech enhancement and suppression method for multichannel auditory signals recorded with both air-and body-conductive microphones. In processing auditory sound signals recorded with wearable microphones for memorizing conversation and acoustic scenes and events that each user experienced in daily life, source separation is a promising technique as only a mixing of various sound sources can be observed. To separate the recorded signals into self-produced speech, i.e.,user's own speech and ambient environmental sounds, we propose a self-produced speech enhancement and suppression method using not only air-conductive microphones to record multichannel air-conducted signals but also body-conductive microphones to record body-conducted signals. A separation filter applied to the multichannel air-conducted signals is estimated in an unsupervised manner while effectively using the body-conducted signals dominantly including self-produced speech components. We conduct an experimental evaluation to investigate the effectiveness of the proposed method, thereby demonstrating that the proposed method outperforms the conventional method using only the air-conductive microphones.