Distant-talking speaker identification using a reverberation model with various artificial room impulse responses

In this paper, we propose a distant-talking speaker recognition method using a reverberation model with various artificial room impulse responses. These artificial room impulse responses with different speaker and microphone positions, room sizes, and reflection coefficients of walls and convoluted with clean speech are used to train an artificial reverberation speaker model. This artificial reverberation model is also combined with a reverberation speaker model trained with room impulse responses measured in real environments. Speaker identification performance using a combination of the two reverberation speaker models achieved a relative error reduction rate of 50.0% and 78.4% compared with that using a reverberation model trained with real-world room impulse responses and a clean speech model, respectively.