1.   The speech output appears rather slow right now. The reason for this is that basic sounds have been recorded individually instead of being recorded as parts of words. The next version aims to do precisely this. This should speed up the speech output and also reduce the size of the database down to about 500KB.

2.  Dynamic speed and pitch modification needs to be incorporated. We have TD-PSOLA code running on a prototype and could add that to the next version, but somebody please tell us whether this method is patented and therefore, not amenable to free distribution.

3.  Consonant-Consonant junctions are currently not  very satisfactory; either there are large gaps, or sounds get gobbled up. Some attention needs to be paid to this. The next version will record more of these cluster characters.

4.  Multiple voices need to be added. A major challenge here is to automate the process of going from recording to setting up the database. The main obstacle here is that of automatic segmentation of the recording. Work is currently in progress on this front.

5.  The phonetic description used by this version is cumbersome; the next version needs a more friendly phonetic description, something which is akin to transcription in Roman, possibly with markups for duration and pitch.

6.  Text-to-Phonetics modules need to be written for all Indian languages. Currently we have only Hindi and Kannada, all other languages need to be addressed.

7.  As mentioned on the technical description page  the Hindi text-to-phonetics algorithm for compound words needs to be improved, i.e., compound words need to be identified,  decomposed into basic words, and the algorithm must be run on each basic word. It is not clear how this can be done without using a lexicon.

If you can help on any of the above, contact Ramesh Hariharan(ramesh@csa.iisc.ernet.in)

Inquiries & questions to <simputer AT csa dot iisc dot ernet> dot in.
design:corrosive

This page maintained by <manohar AT csa dot iisc dot ernet dot in> with scripts and support from <vinay AT csa dot iisc dot ernet dot in>