Sanjeev Satheesh (Baidu, Inc.) “Bias Reduction in Production Speech Systems”
3400 N Charles St
Baltimore, MD 21218
USA
Abstract
Deep learning has helped speech systems surpass humans on speech recognition tasks for multiple languages. One could say, therefore, that the automatic speech recognition (ASR) task may be considered “solved” for any domain where there is enough training data. However, production requirements such as supporting streaming inference bring in constraints that dramatically degrade the performance of model – typically because models trained under these constraints are in a high bias regime and can no longer fit the training data as well. In this presentation, we talk about building deployable model architectures with low bias. This is important because low bias models enable us to:
1)Serve the very best speech models and
2)Build a single speech recognition system that is super human on a wide variety of domains, by adding more data and parameters
Biography
Sanjeev has been a deep learning researcher, and is currently leading the speech team at the Silicon Valley AI Lab at Baidu USA. SVAIL has been focused on the mission of using hard AI technologies to impact hundreds of millions of users. Sanjeev has a masters’ degree from Stanford, where he worked with Fei-Fei Li and Andrew Ng.