Improving Voice Assistants’ Understanding Through Joint Contextual Learning
The quality of automatic speech recognition (ASR) is critical to AI Assistants as ASR errors propagate to and directly impact downstream tasks such as language understanding (LU) and dialog management. In this talk, I will go over multi-task neural approaches to perform contextual language correction on ASR outputs jointly with LU to improve the performance of both tasks simultaneously. I will share the results obtained using state-of-the-art Generalized Pre-training (GPT) Language Models based joint ASR correction and language understanding tasks.
Sai is part of the conversational AI team at Uber, working on building conversational agents for Uber partners. In addition, he is also working on Ludwig, a code free deep learning toolbox open sourced by Uber AI. Prior to joining Uber AI, he led Uber’s efforts in reducing payment losses and Risk support costs using Machine Learning models.
Prior to Uber, Sai was a Master’s student in the Language Technologies Institute at CMU working with Prof. Alan Black on developing novel methods for Speech Translation in zero resource languages. Before CMU, he worked at Microsoft Research on developing automatic transcription techniques for Indian classical Music. He holds a degree in Electrical Engineering from IIT Kharagpur, where he worked on Digital Signal and Speech Processing problems.