 |
International Workshop on Hands-Free Speech Communication (HSC2001)April 9-11, 2001 |
 |
|
International Workshop on Hands-Free Speech Communication (HSC2001)April 9-11, 2001 |
|
We have developed a method of segregating desired speech from concurrent sounds received by two microphones. This method, called SAFIA, uses two microphones and calculates differences in the amplitude and phase between the channels. These differences are used to select the frequency components of the signal that come from the desired direction and to reconstmct these components as the desired source signal. To clarify the effect of frequency resolution on SAFIA, we analyzed the relationship between the frequency resolution and the power spectrum's cumulative distribution. We found that when the frequency resolution was about 10 to 20 Hz, the speech-signal power was concentrated In specific frequency components. Finally, we demonstrated that SAFIA improves the recognition rate of speech signals interfered with by other speech. By applying SAFIA, the error reduction rates of the male and female voices became 76.1% and 75.0%, respectively.
Bibliographic reference. Aoki, Mariko / Furuya, Ken'ichi / Yoshioka, Osamu / Yamaguchi, Yoshikazu (2001): "Sound source segregation based on estimating incident angle for hands-free speech recognition", In HSC2001, 47-50.