Bathina Technologies India Ltd has claimed a breakthrough in voice automation technology by creating voice models for users with different accents with a training (by reading one script) of only 10 to 15 minutes, yielding more than 92 per cent initial recognition accuracy.
Reading additional stories and continued usage of the product improves the recognition accuracy to higher levels of 98 per cent - 100 per cent, the company said in a notice to the Bombay Stock Exchange today.
Generally, most continuous speech recognitions engines come with generic pre-defined speaker-independent voice models. With so many accents in India and around the world, such models give moderate recognition accuracy to majority of the users.
Speech engine developers also commonly provide mechanisms to modify the generic voice-model to create user