An HMM-based speech synthesis system applied to Arabic

Abstract EN

A number of issues related to speech-synthesis with Hidden Markov Models (HMM) are discussed.

The HMM as a suitable model for time sequence modeling is used for estimation of speech synthesis parameters, A parameter sequence is generated from HMMs themselves whose observation vectors consist of spectral parameter vector and its dynamic feature vectors.

HMMs generate cepstral coefficients and pitch parameter which are then fed to speech synthesis filter named Mel Log Spectral Approximation (MLSA).This paper explains how this approach can be applied to the Arabic language to produce intelligent Arabic speech synthesis as in the diphone concatenation.