| 摘要 |
To manipulate an audio signal, a first overlapping chain of windows is generated, successive windows being placed incrementally, each placed a pitch period after its predecessor. In each window, the signal is weighted, and this yields a signal segment for each window. The segments are subsequently placed in a second overlapping chain, in which the segment positions are modified as compared to the first chain, some segments being repeated or skipped. When the segments thus placed are summed, this produces a high quality signal with pitch and/or duration changed with respect to the original signal. The invention is used amongst others for diphone speech synthesis, the relative positions of the diphones moreover being adjusted to minimize audible transition effects between diphones. In an embodiment, the audio signal used as input is first manipulated to give it a monotonous pitch, and later manipulated a second time to give it a pitch with a desired temporal variation in pitch and/or duration. <IMAGE> |
