Wav2Letter++: A Fast Open-source Speech Recognition System

By Vineel Pratap, Awni Y. Hannun, Qiantong Xu, Jeff Cai, Jacob Kahn, Gabriel Synnaeve, Vitaliy Liptchinsky, Ronan Collobert. 2018

168 citations recorded by the source corpus.

This paper introduces wav2letter++, a fast open-source deep learning speech recognition framework. wav2letter++ is written entirely in C++, and uses the ArrayFire tensor library for maximum efficiency. We explain the architecture and design of the wav2letter++ system and compare it to other major open-source speech recognition systems. In some cases wav2letter++ is more than 2× faster than other optimized frameworks for training end-to-end neural networks for speech recognition. We also show that wav2letter++ training times scale linearly to 64 GPUs, the most we tested, for models with 100 million parameters. High-performance frameworks enable fast iteration, which is often a crucial factor in successful research and model tuning on new datasets and tasks.

Subfields: Speech & Audio

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Speech & Audio

Citations in the corpus

wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations (2020, methodology): “We use the beam search decoder of [44].”
Unsupervised Speech Recognition (2021, methodology): “For Librispeech experiments, we also decode with a Transformer language model (Baevski and Auli, 2018) trained on the Librispeech LM corpus using the beam search decoder of Pratap et al. (2019).”
Robust wav2vec 2.0: Analyzing Domain Shift in Self-Supervised Pre-Training (2021, methodology): “Wav2letter++ [42] beam search decoder are used with a beam size 50, beam threshold 100, and Bayesian optimization2 is used to find decoding hyper-parameters over 100 trials: LM weight ([0, 8]), word score ([−5, 5]), and silence score ([−5, 5]).”
MLS: A Large-Scale Multilingual Dataset for Speech Research (2020, methodology): “All the baseline models are trained using wav2letter++ [19] framework.”
Self-Training and Pre-Training are Complementary for Speech Recognition (2020, positive): “(2020; [2]) and train a Transformerbased sequence to sequence model with log-Mel filterbank inputs after pseudo-labeling using wav2letter++ [37].”
Rethinking Evaluation in ASR: Are Our Models Robust Enough? (2020, methodology): “All experiments are implemented within wav2letter++ [Pratap et al., 2019].”
Massively Multilingual ASR: 50 Languages, 1 Model, 1 Billion Parameters (2020, methodology): “All our experiments are run using wav2letter++ [17] framework.”
Libri-Light: A Benchmark for ASR with Limited or No Supervision (2019, positive): “We then ran a Voice Activity Detection (VAD) model using the wav2letter++ framework [18] on the recordings to tag onsets and offsets of speech segments.”
Effectiveness of self-supervised pre-training for speech recognition (2019, methodology): “We use the publicly available wav2letter++ (Pratap et al., 2019) decoder integrated into the Fairseq framework with the ofﬁcial Librispeech 4-gram language model.”
Self-Training for End-to-End Speech Recognition (2019, methodology): “Experiments are done in the wav2letter++ framework [17].”
WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing (2021, methodology): “For evaluation, we use wav2letter++ [88] beam search decoder with language model (LM) fused decoding as :”
SUPERB: Speech processing Universal PERformance Benchmark (2021, methodology): “The trained model is decoded with LibriSpeech ofﬁcial 4-gram LM powered by KenLM [37] and ﬂashlight [38] toolkit.”