GTC 2020 S21263
Presenters: Vitaly Lavrukhin ,NVIDIA; Jocelyn Huang, NVIDIA
Abstract
We’ll present our latest automatic speech recognition models that reach a state-of-the-art accuracy while having almost 10x fewer parameters than Jasper, our previous flagship model. The small model size enables deployment on a broad spectrum of GPU-accelerated devices, from DGX servers all the way up to tiny Jetson Nano. We’ll give architecture details and training recipes in NeMo. Finally, we’ll discuss different approaches to transfer learning and domain adaptation.
Watch this session
Join in the conversation below.