| IT Product Re-engineering
Project
IBM US/UK English ViaVoice to Indian English ViaVoice
Description
To adapt IBM ViaVoice UK/US continuous speech dictation system to the
Indian English accent. This involves adding Indian words, names and
incorporating the Indian style of speaking into the existing system.
At present the speech dictation products offered in the market are not
capable of recognizing the Indian accent correctly, the Indian dialect
and way of pronunciation.
The software was developed in the AIX environment; it was then migrated
to the window environment. The software will work with Windows95/98/2000/NT workstation.
SST developed the language model, acoustic model and DLLs required for the speech
engines in the software. The acoustic model of the software was framed after analyzing
about 40,000 sentences spoken by the geographically distributed Indian speakers.
The starter set vocabulary was framed with 62,000 words occurring frequently in conversations.
For this purpose 16 million sentences were collected from various Indian publications and
analyzed and the proper nouns added to the vocabulary set. The software has been trained
to recognize words such as lakh, crore, Mumbai, Chennai, Delhi and other proper nouns that
are unique to India.
The result was a fully operational speech dictation software that included words
used on an everyday basis in many environments - the home, the workplace, the school,
etc. Add to this the provision to include another 62,000 personal words - a strong
vocabulary base to understand whatever you say. The words in ViaVoice vocabulary can be segregated to
base vocabulary, personal vocabulary and active or selected topics. ViaVoice works by looking
for the words in all the three vocabulary data banks and if it does not find it there, then it
looks up in the backup vocabulary and adds it to the personal vocabulary.
Top
Project
Testing of Speech Runtime APIs.
Description
The project involved developing test plan and test cases that are repeatable and then
execute them on multiple platforms, for speech run time APIs. The APIs are for speech
recognition and Text-to-speech (TTS) synthesis. The run times are across platform including
Windows 95/98/NT, Linux and AIX. The speech applications covered were desktop, embedded
and telephony. The test cases were developed on Java platform.
Top
Project
HANS (HDSL Access Network System)
Description
The HANS (HDSL Access Network System) project done for Crompton Greaves involved developing
3 kinds of Systems.
HANS for E1 transport.
HANS for Pair-gain Application.
HANS for NX64 Leased Line Application.
HANS for E1 transport
A Card was developed which translated E1 format to HDSL at an exchange side and back
to E1 format at the remote side. Number of such cards can be stacked in a Back-plane
in order to support many E1 Streams. NMS support was provided using RS-232 between PC and
one Card, using RS-485 between cards.
HANS for Pair-gain Application
This involved translating 30 analog voice channels to HDSL format at Exchange side and
back to analog format at the remote end. NMS support was provided using RS-232 between PC
and one card and using RS-485 between cards.
HANS for NX64 Leased Line Application
This system supported interfaces like V.11, V.35, and V.36 with varying
rates (from 320kbps to 2.048Mbps in steps of 64kbps). Top
Project
Develop a 2-Channel CDMA Base-Band Receiver
Description
The 2-channel CDMA Base-band Receiver developed along with MIT, is a part of a CDMA system. The base-band card was responsible for IF to Base-band translation and extraction of data (after code and carrier recovery).
The input to this card is an 8MHz IF signal (derived from a RF Card). The Card has a
SHARC (ADSP-21062) along with two Actel FPGA's (each for one channel) to implement the
tracking algorithm. The tracking algorithm is a combination of Costas Loop (for Carrier tracking)
and Delay Locked Loop (for Code tracking). The card then transfers the data stream to the
Back-plane in MIL-1553B format
Top
|
