AI - Sundeep Teki

ARTIFICIAL INTELLIGENCE

My work is focused on applications of deep learning for improving virtual assistant AI using:

End-to-end Speech Recognition for Virtual Assistants

Developed state-of-the-art end-to-end deep learning ASR models based on sequence-to-sequence models
Significantly improved ASR model training times using ring-allreduce distributed training
Trained ASR models on >20000 hours of data using cloud infrastructure

NLP: Sensitive interactions between Humans and Virtual Assistants

Trained deep learning NLP models to detect sensitive content in human-assistant interactions
Deployed NLP models for non-English languages using Neural Machine Translation and Text Generation techniques

Privacy-preserving deep learning

Researched homomorphic encryption, federated learning and its applications for private machine and deep learning

Speech Recognition for Low-resource, Code-Mixed Indian Languages

ASR modeling for "Hinglish", i.e. Hindi interspersed with English common amongst Indian speakers
Spoken Language Identification for low-resource Indian languages code-mixed with English
- 2nd in Microsoft Challenge on Speech Technologies for Code-Mixing; Paper accepted at INTERSPEECH 2020
Speaker Diarization modeling, i.e. "who spoke when" for code-mixed speech
Audio Sentiment Analysis to understand emotions expressed in speech

Applied Natural Language Understanding