Sreedhar Patha Email & Phone Number

Currently working on ASR for health domain predominently.My previous work is related to ASR (Automatic Speech Recognition). Trying to build ASR model for English from different accents and a universal ASR model for similar family of Indian Languages. Have built many domain specific ASR systems for various domains (Travel, BFSI...) and languages (UK, US, Indian English, other Indic Languages).Handled various adaptation requirements including incorporating different variables like contact numbers, names etc on the fly, phone level decoding for the unknown words in the system, Language model (LM) adaptation, Acoustic model (AM) adaptation, Byte pair encoding (BPE) based lexicon and on the fly graph composition.Built Pronunciation models (PM) for various Indic languages at different levels viz., phone, syllable and morpheme.Exploring new unsupervised/ semi supervised methods of collecting data and developing an open data set in Indian languages. Building an entire pipeline to collect, decode and test in a semiautomatic fashion.My research work is related to importance of syllables in speech signal analysis.A new way of extracting syllables has been proposed, studies have been conducted to see the efficiency of this approach in improving ASR accuracy.A small work has been done on usage of syllables in audio search which gave positive result.Speaking rate estimation using these syllables and incorporated it in ASR to improve accuracy.An effort has been made to reduce WER of ASR for Indian Languages by mapping various similar sounding phones at syllable level.Currently imroving on extraction of syllables method and automatic incorporating of speaking rate in ASR.

Listed skills include C, Linux, Matlab, Python, and 4 others.