MBROLA:
A speech synthesizer program.
The multilingual applications
developed at IIT Madras, use the MBROLA speech synthesizer program to produce
speech output from the Indian language text dealt with by the application.
The applications include
1. Multilingual editor
2. A text reading program
to read Indian language text.
3. Sound enhanced web browser
(based on Lynx)
4. An interface program
to work with Jaws for Dos under Win95.
All these applications
cater to the multilingual requirements within the country. The choice of
MBROLA as the speech engine was influenced by many factors. the development
team at IIT Madras found this to be an excellent piece of software that
relates well to the phonetic nature of Indian languages by allowing
phonemes to be synthesized elegantly. The syllable level representation
of the text in Indian languages, a feature unique to the IIT Madras
software, makes it very easy for applications to support speech output.
The Mbrola system
is not an application that is ready for use. It is a building block in
a system for producing speech and hence one must use it in a computer program
which can identify the phonemes to be used for speaking out a text string.
The MBROLA development team has been generous in making the software available
as a library of functions which could be called from a computer application.
They have this library available for a number of computer systems including
Linux, Windows, Dos etc., and hence one can develop speech output applications
on almost all the computer systems in use today.
MBROLA derives its
basic strength from the fact that it is a piece of software that produces
speech output from a list of phonemes. Thus, its operation is not specific
to a language and one can get true multilingual output using the speech
engine by feeding appropriate phonemes to the engine and indicating the
language used for generating the phonemes. Mbrola achieves this excellent
capability through the concept of a speech data base which has the information
required for producing the sound associated with the phoneme. The Mbrola
team has been assisted by volunteers from many countries to produce speech
databases in many languages of the world. Also, volunteers have contributed
programs which could be used for mapping text strings in different languages
to the associated phonemes thus making the job of text to speech conversion
relatively easy.
In respect of Indian
languages, suitable data bases have not been developed yet and may take
some time. However, it is possible to produce very satisfactory results
for Indian languages, by using phonemes of languages for which a data base
is already available. We have observed that Swedish has most of the phonemes
required for the Indian languages and hence the development team at IIT
Madras has used the same for its applications.
(Note: As of 2004, Mbrola
voice data bases for Hindi and Telugu have been created. However, the support
required for a full set of phonemes to cater to all the aksharas have not
been included. A proper voice data base for Indian languages is yet to
be fully realized.)
The procedure for
developing a speech data base for MBROLA, requires that a representative
speech sample for each language be collected (the specs for this are available),
a recording made and from this identification of diphones (the continuation
of one phoneme into another in a smooth fashion) completed. For lack of
time and recording resources, the IIT Madras development team has put this
exercise off for the time being. Persons interested in contributing to
MBROLA speech data bases for Indian languages may kindly get in touch with
the IIT Madras development team via email (details in the contact
IIT page).
The data base used
by the IITM team for speech output in English is known as EN1. The speech
produced is very typical of what one might hear in England. For those in
India, this is closer to their own understanding of spoken English compared
to English spoken in the U.S.A.
IIT Madras finds that
MBROLA is very easy to work with on Windows as well as Linux platforms.
Downloading and installing the software is also straightforward but help
from a sighted friend may be needed to complete the installation on both
systems.
