To appropriately discriminate and separate voices even in the case of the tone of fundamental frequency of mixed voice being in proximity by extracting voiced sound parts individually while adding information of sound source direction of voiced sound, and supplementing individually extracted voiced sound groups with voiceless sound to extract voice signals.
A voiced sound extracting part 103 extracts voiced sound parts of voice individually from an input acoustic signals of acoustic input terminals 101, 102. A sound source orientation part 104 extracts the sound source azimuth of each voiced sound. A voiced sound grouping part 106 extracts individual voiced sound, dividing them in voiced sound groups speaker by speaker and outputs them along with the direction (d) of the sound source. A residual extracting part 105 subtracts the waveform of all voiced sound from mixed acoustic input signal waveform inputted from the terminals 101, 102 and outputs residuals. A voiceless sound extracting part 108 extracts acoustic components in each sound source direction in the residuals and outputs them. A voiceless sound supplementing part 107 adds voiceless sound waveform to the waveform of each voiced sound group to extract each voice.
OKUNO HIROSHI
KAWABATA TAKESHI
Next Patent: VOICE RECOGNIZING DEVICE