Indian Language Speech Recognition

This technology is a Speaker Independent Recognition System for Large Vocabulary. It can recognize Continuous Hindi/Bengali Speech and can also convert Speech to Text. It also enables Speech based Human Computer Interaction. It can be used for Call center data transcription .It can also be used for Speech based enquiry system which can be employed in services like railway reservation. The following algorithms have been used in the technology

• MFCC features
• Hidden Markov Models
• Bigram language models

It will open up a number of Indian language applications.

• The technology has the capability to recognize Hindi speech. The IBM system is proprietary and has a good accuracy but it is also costly.
• This is the only system that can convert Bengali speech to text.
• Unlike existing English ASR systems, it can be trained on Indian languages.
• It consists of a novel triphone clustering scheme that captures context information with higher accuracy and less training.

Triphone HMM Word Level Accuracy