Hello there!
I’m Shreekantha (Shree). I’m currently a Senior Speech Recognition Engineer at Dialpad Canada, leading the next-gen ASR product efforts. Before this, I was pursuing MS by Research in Data Science at IIIT-Bangalore under the supervision of Prof. Sachit Rao and Prof. V. Ramasubramanian, where I also worked at the E-Health Research Center (EHRC) and the Machine Intelligence and Robotics Center (MINRO) as a Research Scholar. I graduated with a thesis on “Multi-task learning in end-to-end attention-based automatic speech recognition”. Before joining IIIT-Bangalore, I was with Sonus Networks (Now Ribbon Communications) where I worked on developing and testing Element Management Systems for 4G-VOIP products. I have a Bachelor’s degree in Telecommunication Engineering from JNNCE Shivamogga (VTU).
My research interests include:
- Streaming end-to-end ASR for conversational, telephony, and videoconferencing speech
- Low-latency and computationally constrained scenarios
- Multi-lingual and code-switched speech recognition
- Bringing external knowledge into the purely data-driven end-to-end architectures
When I’m not building next-gen ASR products, I conduct research to bring in some classical speech knowledge to the purely data-driven models of recent years. Through this exercise, I’m hoping to blend the pure data-driven architectures with speech knowledge, leading to a reduction in model complexity, faster training/inference, and hopefully, deeper insights into speech recognition.
Other problems I am currently working on include
- KWS using neural attention
- A pre-training method for end-to-end ASR
- Better training strategies for encoder-attention-decoder models
- Interpretability and explainability of end-to-end ASR models
- Multilingual and code-switching scenarios
- Gathering data and building ASR models for low-resource Indian Languages
- Kannada
- Sanskrit
Most of my work is in kaldi/K2, ESPnet, NeMo, and some custom code in PyTorch ❤️ and TensorFlow 2^.
You can reach me at shreekantha.nadig@iiitb.ac.in