|
|
 |  |
 |
| Title | Speech Information Processing - Development of Algorithms and Technologies |
| Time Frame | One year 1st February 99 to 31st January 2000 |
| Sponsoring Organization | Dr. Sira G. Rao,
M/s. BPL Telecom Ltd.,
No. 1/1 Palace Road |
| Project Executive | Prof. T V Sreenivas Dept. of ECE. |
SCOPE OF WORK
Institute shall undertake the project to design algorithms for the purpose of developing the voice dialer. The specifications of the voice dialer are;
- Isolated words recognition of limited vocabulary
- Vocabulary size of approximately 20 English words. Consisting of digits and command words.
- Recognition accuracy better than 95% with toll quality speech.
ACHIEVEMENTS
It was felt that speech is the most natural form human communication. The goal of automatic speech recognition is to develop a system which will enable humans to interact with machines using such a natural mode of communication. The problem of speech recognition was broadly classified into two categories based on the type of models used for the recognition unit (word) (i) Non- parametric model and (ii) statistical parametric model. The project aimed at designing a voice activated telephone dialing system for a desktop application. It was understood that a voice dialer requires a small size vocabulary and it should accept isolated word mode of speaking. Hidden Markov Model (HMM) was chosen as the most suited one for speech modelling and recognition. All algorithms were developed in high level DSP programming environment, MATLAB and later converted to C++ executable code under linux operating system. This enables to construct any required combination of the m. files into an executable (for a linux - pc) and make it available.
Extensive work and tests were carried out on speaker dependant multi speaker ASR System- I and System - II.
The final result has been passed on to BPL.
|
|
|
