Sri Lankan English Voice for Text to Speech Applications

Abstract

Speech is the main communication method used to convey ideas in daily life. Text to speech is the conversion of written text to spoken speech. There are many software and hardware systems available for text to speech process, but these are having limited number of voices mainly for languages like US English, UK English and Spanish. Additional voices available for these are not easily obtained while there is an extra cost involved. So it is difficult to find various sounding voices for the ordinary end user of speech synthesis systems. There is no synthetic voice available for the Sri Lankan style English. Customizing a voice for a particular target user is still in a research stage. This research project was focused on the above-mentioned problems while attempting to give feasible solutions. The project was done for a voice synthesis engine called Festival. As an outcome of this project a new synthetic Sri Lankan English voice was created. Also further research was done on voice customization for a particular speaker by implementing a voice-building interface for end users to customize voices easily. The generation of the Sri Lankan English voice was done using „diphone concatenation synthesis‟ technique. Voice conversion technique was used for the voice generation interface. The created Sri Lankan English voice was in an average level of quality, which can be improved further. The voice customized from the voice-building interface showed a good level of quality although the similarity between the target speaker‟s actual voice and synthetic voice is only average. Providing the voice-customization facility to the end user was proved to be practical and needs more attention since currently this facility is not provided to the end user in most text to speech systems.