Software for scanning of Indian rarities?

Claude Setzer cssetzer at MUM.EDU
Thu Mar 19 20:03:24 UTC 1998

Good points, but remember that voice recognition coupled with powerful
artificial intelligence is very trainable....far more trainable I think than
the best text recognition systems that I have seen. Also the human voice can
be far more consistent than the many font faces available....almost
unlimited. Even in English using the best of quality fonts, can be beyond
the capability of the best OCR. Scanning a real book is another problem
altogether. The speaker can compensate for many of the problems of the
printed text.

I don't know how soon this will be available but I believe that voice
recognition has more potential than OCR.
I also am not advocating even the possibility of universal untrained voice
recognition, but the use of a carefully trained and practiced human speaker.
There is really no possibility to "train" a specific text, especially an old
one. With OCR you can train only the software, not the "speaker."

Claude Setzer

-----Original Message-----
From: Viktor V. Sukliyan <madhava at CH1.VSNL.NET.IN>
Date: Thursday, March 19, 1998 1:30 PM
Subject: Re: Software for scanning of Indian rarities?

>On Sun, 15 Mar 1998, Claude Setzer wrote:
>> Another option which may only a slight chance of being implemented is the
>> Verbot (PC base verbal robot) approach which carries on conversations in
>> English now and promises to be extended to other languages in the next
>> years. I think these people are going to have better and less expensive
>> voice recognition than what is available now, and since Sanskritists  are
>> sometimes willing to put more attention on "consistent" pronunciation
>> speakers of other languages, this might prove to be quite interesting as
>> way to enter text.
>> Claude Setzer cssetzer at
>I don't think it's practically possible. It won't be possible for robot to
>tetermine accoustical difference appearing in pronounciation
>of different speakers, say from Bengal (can't pronounce such consonants
>as v, ya), Punjab (instead of krishna-krishan, dharma-dharam etc),Tamils
>(I suspect some difficulties with ka, kha, ga, gha, as it was mention on
>Indology previously ), English speakers are in more difficulties with
>their specific phonetic system situated very far away from that of
>Secondly voice recognition is not so universal as optical one where you
> need not to have phonology experts qualified to read and speak.

More information about the INDOLOGY mailing list