Espresso: A Fast End-to-end Neural Speech Recognition Toolkit
This provides an efficient and modular toolkit for researchers and practitioners in automatic speech recognition, though it is incremental as it builds on existing frameworks like PyTorch and fairseq.
The authors tackled the problem of building a fast and high-performance end-to-end neural speech recognition toolkit, resulting in Espresso, which achieves state-of-the-art ASR performance on datasets like WSJ, LibriSpeech, and Switchboard without data augmentation and is 4-11 times faster for decoding than similar systems.
We present 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 data sets among other end-to-end systems without data augmentation, and is 4--11x faster for decoding than similar systems (e.g. ESPnet).