GTC 2020 S21239
Presenters: Yiming Wang,Johns Hopkins University
We’ll introduce Espresso, an open-source, modular, extensible, end-to-end neural automatic speech recognition (ASR) toolkit based on the deep learning library PyTorch and the popular neural machine translation toolkit fairseq. Espresso supports distributed training across GPUs and computing nodes, and features various decoding approaches commonly employed in ASR, including look-ahead word-based language model fusion, for which a fast, parallelized decoder is implemented. Espresso achieves state-of-the-art ASR performance on the WSJ, LibriSpeech, and Switchboard datasets, among other end-to-end systems without data augmentation, and is up to 11x faster for decoding than similar systems, such as ESPnet.
Watch this session
Join in the conversation below.