잡음에 강인한 음원 방향 검지 시스템을 위한 잡음제거와 음성검출기에 관한 분석
A study on speech enhancement and voice activity detection for robust sound source localization in noisy environments
음원 방향 검지 잡음제거 Voice Activity Detection;
- 원문 URL
Human-robot interaction or human-robot communication based on audio techniques has been continuously studied for a long time. However, the sound system has been recently applied to appliances in real life. And the performance is not good in real noisy environment, although the sound system can be conducted well in quiet environments. So, this paper proposes to detect the direction of a person (speaker) who is calling the cleaning robot while it is operating. In addition, this study has the possibility to be transplanted to other appliances, because it is handling heavy noises the cleaning robot makes during the vacuum operation. The lowest level of the SNR (Signal to Noise Ratio) in this experiment is -7dB. For this implementation, Speech Distortion Weighted Multi-channel Wiener filter (SDW-MWF) and Karhunen-Loeve Transform (KLT) based filter are used to remove the cleaning robot noises. And a voice activity detection (VAD) technique using the Band-pass filter and the harmonic characteristics estimates the frames including human voices. In addition, Phase Difference of Arrival (PDOA) is applied to find out the direction of speaker. Finally, the result of this experiment presents the enhanced performance of the sound source localization in corrupted environments.