Wav2Letter: an End-to-End ConvNet-based Speech Recognition System

TL;DR

This paper introduces Wav2Letter, an end-to-end convolutional neural network for speech recognition that simplifies training and achieves competitive word error rates on Librispeech, using raw waveform and MFCC features.

Contribution

It presents a novel end-to-end speech recognition model with a new segmentation criterion that simplifies training without requiring phoneme alignment.

Findings

Competitive word error rate on Librispeech with MFCC features

Promising results using raw waveform input

Simpler training process comparable to CTC

Abstract

This paper presents a simple end-to-end model for speech recognition, combining a convolutional network based acoustic model and a graph decoding. It is trained to output letters, with transcribed speech, without the need for force alignment of phonemes. We introduce an automatic segmentation criterion for training from sequence annotation without alignment that is on par with CTC while being simpler. We show competitive results in word error rate on the Librispeech corpus with MFCC features, and promising results from raw waveform.