Human conversation signal recorded in a real environment enevitably contains not only target speaker's voice but also noise, reverberation, and other speakers' voice, which are known to degrade speech recognition performance of the target speaker. I investigate denoising, dereverberation and source separation technologies for removing such interferences.