Method and apparatus for speech recognition.

摘要

The processor includes a device which linearly increases an input speech signal (A) frequency by a factor of 2 or 3. An excitation signal (E) representing word sound and vibration is extracted from the speech signal. An speech signal spectral envelope allows speech signal coefficient characteristics to be determined. An interpolation device (10) calculates an interpolated excitation signal from the excitation. The interpolated excitation signal has a timing diagram signal which is identical to the timing diagram of the excitation signal. The density of values in the interpolated signal is two or three times the density of values in the excitation signal. A more intelligible speech signal is synthesised from the interpolated excitation signal and the characteristic coefficients.