This paper proposes a new speech coding method called PSI-CELP (Pitch
Synchronous Innovation CELP). This method is based on CELP, but adds p
itch synchronous innovation, which means that even random codevectors
are adaptively converted to have pitch periodicity for voiced frames.
In addition, this paper also proposes four other methods: (a) a fixed
codebook that is prepared as part of the adapative codebook for nonper
iodic frames, (b) a random codebook which consists of two conjugate su
b-codebooks, (c) an MA-type perceptual weighting filter using unquanti
zed LPC parameters, and (d) codebook searches using delayed decision.
PSI-CELP improves the SNR, segmental SNR, and opinion equivalent Q val
ue of the coding system by 1.6 dB, 1.0 dB, and 0.5 dB respectively.