To alter the pitch in the voiced speech case, an LPC analysis and
reconstruction method is used, Figure
F0 modification is achieved by exciting the LPC model with a duration
adjusted residual. The duration of the residual is adjusted by
windowing, and for the case of decreased pitch, zero padding. The
window applied to the residual is defined by:
where T is the desired output pitch period duration, or the full
duration of the original (input) pitch period, whichever is smaller.
For increased F0 (T less than original pitch period duration), the
windowed (and shortened) residual is used to excite the synthesis
filter. To decrease F0, the windowed residual error is zero padded to
the desired duration and then used to excite the synthesis filter.
The pitch in voiced speech is thus modified by altering the duration
of the LPC model residual error and exciting the model with the
altered residual. Additionally, the LPC parameters and residual are
inter-frame smoothed through linear interpolation during the overlap
interval prior to the nominal pitch period onset. This smoothing
functions to decrease ringing and other artifacts at pitch period
boundaries.
Figure 1:
Block diagram of HYBRID-LPC pitch modifier for the case of voiced speech.
Next: Listening Test
Up: Hybrid LPC Method
Previous: Hybrid LPC Method
Tim Bunnell
Tue Dec 5 10:53:09 EST 1995