Multipulse is a well known, analysis by synthesis technique to model the LPC residual signal , With this model, speech can be coded at medium bit rates with good quality. However, the determination of the pulse locations is a combinatorial problem with no closed form solution. For these reason suboptimal procedures are used. In this paper, inhomogeneous SA structures have been used for solving the multipulse problem; the obtained performances are compared with suboptimal algorithms such as the Atal's ones. As compared to the one described in , this algorithm achieves an effective improvement in objective performances. As a matter of fact, an improvement of more than 1.5 dB is obtained. A set of performance curves will be also described in order to find the optimum trade-off configurations.
Keywords: Stochastic Optimization, Simulated Annealing, Speech Coding, Multipulse Models
Bibliographic reference. Mumolo, Enzo / Rebelli, Alessio (1993): "Optimal multepulse excitation determination by simulated annealing", In EUROSPEECH'93, 1131-1134.