摘要 |
<p>PURPOSE:To recognize the most stable phonemes in real time by extracting the phonemes from a continuous voice. CONSTITUTION:The voice is inputted to a linear Fourier transformation part 11 through a microphone 16 and an AGC amplifier 17. Images appearing at spatial modulators 23-25 are exposed to light from a laser diode 1 which has a uniform wave front under the control of LCD control circuits 20-22, and images generated by processing framed voice signals into a power spectrum, quefency, and a 2nd power spectrum by Fourier transformation are photodetected by photodetection parts 26-28. A difference circuit 41 compares the 2nd framed power spectrum at a current point with that at a next point and also compares the power spectrum in a next stage with that in a further next stage to extract the framed power spectrum having a minimum value. This spectrum is developed in two dimensions to form a phoneme pattern through a spatial modulator 43 and a photodetection part 45, and this phoneme pattern is compared by a phoneme recognition part 14 with phoneme patterns which are already registered to recognize the phonemes.</p> |