I am from Sri Lanka. We are creating a virtual avatar that can speak English, Tamil and Sinhala. I tried the eSpeak Sinhala and Tamil voices. They are not very clear and are robotic. We want to create new voices for Sinhala and Tamil for eSpeak while the list and rules remains the same. How can we create this voice file? How to integrate it with eSpeak?

Assume we have the resources to create a new voice. 

Any espeak-ng voice will sound more or less robotic.
But, for example, English is quite good even with generated sound
because of elaborate phoneme adjustments for it.
To create your own definitions of espeak-ng sounds look at
https://github.com/espeak-ng/espeak-ng/blob/master/docs/phontab.md
and
https://github.com/valdisvi/espeak-ng-espeakedit

Semi-solution would be using Klatt voice, which is more human
than standard voice. Unfortunatelly there is no known documentation
how to develop voice and use it from espeak-ng.

Another solution is to use MBROLA, which uses recorded diphthongs.
It is free to use, but not open source software. Look for it at
http://tcts.fpms.ac.be/synthesis/mbrola.html

Valdis