Abstract
The paper propose a noise reduction method using observed mixture signals by two microphones. A joint distribution of observed mixture signals has straight lines depending on transfer functions. Therefore, noise signals are reduced by multiplying a weighting factor based on the straight lines to the joint distribution. The proposed method can reduce noise signals even when the number of sound sources is larger than that of observed mixture signals. In addition, our noise reduction method can be processed in real-time. By several simulations, the proposed method can extract the target speech signal from multi speaker's utterance sounds.
