Sunday, September 20, 2009

உரையிலிருந்து பேச்சுக்கு (Text-2-Speech)

Dr. A. G. Ramakarishnanan of IISc posted his Tamil/Kannada Text-to-Speech engine demo link to INFITT Group. This is a test on it with one of my old rambling rants in tamil.

The Original Story is here
பலகணி (Ventilator)

In "Male Machine Reading" is here
Get this widget | Track details | eSnips Social DNA


The engine works charmingly. The main characteristics that I like in it is its speed for the conversion. The wav file is ok with 18 MB for this particular story. Then I used audacity to 'hype' the male (chuvanist) voice and saved as a mp3; 9 mb. Given the size, time for conversion and the quality of the audio file, a nice engine.

Of course, few tunings seem to be necessary in the engine in the future such as
1. short vowels -in times- get chopped off like swallowing the alphebet soup,
2. shrieking vibrating echoes whenever the voice demodulates, and
3. Oh Yes! It is as much as enthusiastic like like Sam Anderson and Ben Stein in reading a story...(That is why it is machine, and it is not a weakness)

With the time, i think this minor issues can be overcome.

Without any question, this is a milestone in Tamil Computing. It will certainly be useful for many people in their personal and professional activities and duties.

Congradualtions to everyone contributing to this project!