Research On Voice Prosody Modification For Mobile And Portable Platforms

Speech prosody modification involves changing the pitch and duration of signal without affecting the naturalness of speech and its spectral envelope. As a branch of speech signal processing, speech prosody modification has important significance theoretically and practically.This paper focuses on improving an algorithm of speech prosody modification, including pitch-scale and time-scale, it is designed to be run on portable devices in real time with good synthesis speech quality. First of all, several typical methods of prosody modification are studied. In the investigation of TD-PSOLA (Time domain pitch synchronous overlap add), some pivotal points, including pitch detection, pitch marking, pitch-synchronous analysis and synthesis, are discussed in details. Secondly, a dynamic programming routine used to calculate an optimal path of a rectangular matrix is introduced in this algorithm for pitch marking. It is discovered that the length of analytic window has different effects on the synthetic signal, i.e. when the length is corresponding to the smaller pitch period, the result is better than the cases when the window is larger. Finally, in order to overcome the discontinued problems of the amplitudes and phases result from waveform copying and deletion by the TD-PSOLA algorithm, the linear prediction technique is introduced in this algorithm. Experimental results show that this improved prosody modification system has good speech quality while keeping lower complexity.In the last part of this paper, a novel algorithm called HNM (Harmonic plus noise model) is addressed. The performance of this model is better than TD-PSOLA, but is higher in complexity. So this model is not suitable for real time application in the current portable hardware conditions.