A Sentence-Pitch-Contour Model for Indiginous Language (Galo) Using Vector Quantization (VQ) and Hidden Markov Model

A model is proposed to developed a Indigenous language (Galo) sentence's pitch-contour with sentence-wide optimization, called the sentence pitch-contour using HMM(Hidden Markov Model) & VQ (vector quantization). To develop a sentence pitch-contour (SPC-HMM), each training sentence are normalized for the pitch-contours of the syllables. Our model is effective for pitch height normalization. After the process of normalization completed, the pitch-contour of each training syllable is vector quantized(VQ). The lexical tones of adjacent syllables and the quantization code are combined both to define the observation symbol sequences for HMM training. Using a dynamic-programming based algorithm in the synthesis phase, for given a sentence and related text-analysis information, the probable observation sequence is generated by finding the sentence wide largest probability path. We have also conducted practical perception tests and is observed that the speech synthesized using the sentence pitch-contour (SPC-HHM) is better than uttered by an ordinary speaker. To get the naturalness of the synthesized speech pitch counter has a great role.