In this paper, a barge-in free spoken dialog system using virtual sound field reproduction is proposed. In the conventional spoken dialog system with an acoustic echo chancellor, it is indispensable to estimate and update the room transfer function, especially when the transfer function is changed by various interference. However the estimation process for the transfer function prevents the user speaking freely and simultaneously with speech responses from the system. In order to resolve the problem, a multi-speaker system is newly introduced to reproduce the virtual sound field that speech responses of the system can be canceled out, however user's speech can be acquired appropriately at the microphone. The speech enhancement and speech recognition experiments reveal that the proposed dialog system has the robustness against changing of room transfer function, and can improve the interference-reduction performance by about 10 dB and the word recognition rate by about 24, in comparison with the conventional system with an echo canceler.
展开▼