Text-to-Speech translation using Support Vector Machine, an approach to find a potential path for human-computer speech synthesizer

摘要

Text-to-Speech (TTS), an astounding feature to assemble computer with intelligence and to induce sound is seemingly a challenging task as it is related to the propagation of uncertainty with the input. This is because TTS evolutes the input based on the probabilities and not with certainty ratios. TTS is accomplished by generating the sound structure/phoneme and then classifying these phonemes in the phonetic dictionary. The Wards' algorithms, BIRCH, Support Vector Machine (SVM) are used to figure out the appropriate sound representation for the given context. To distinguish correct elocution, the SVM procedures are equipped with the principles of pruning. The output was analyzed using divergent stages of uncertainty. In order to study the effect of the output 10 listeners were considered for determining Signal-to-Noise (SNR) ratio. SNR shows that the errors of both type phase and uncertainty were approximately 6% resulting 94% of accuracy. These results manifested that SVM stratagem can be used to obtain better results for TTS synthesizer.