Indian Journal of Science and Technology

Abstract

The speech quality assessment is categorized into two: subjective evaluation and objective evaluation. The subjective evaluation is mainly based upon the perceptual quality of the artificial speech which is carried out by means of listener’s rating. But, here the drastic variation at the points of concatenation is not taken into account. Hence, objective measures are the right alternative for evaluation. Here, the time domain parameters like energy, intonation and duration are analyzed for the synthetic speech. The rating is given on the scale of 1 for the variations and this is deduced from the subjective evaluation results, as the overall quality cannot be purely based on objective measures. The synthetic speech is synthesized by a phoneme based Unit Selection Synthesizer (USS) consisting of three hours speech corpus. The objective evaluation is assigned a value of 1, among which more priority is given to energy as 0.7 in 1 and less priority is given to duration as 0.3 in 1, based on its contribution to the quality and variations. Intonation is not considered, as its characteristics are not revealed at phoneme level. Finally the objective MOS 0.704 is reduced from the actual MOS 2.75 and the overall rating obtained is 2.046.
Keywords: Duration, Intonation, Objective Measures, Short Time Energy, Subjective Measures, Unit Selection Synthesizer