Lipreading with Long Short-Term Memory

TL;DR

This paper demonstrates that a neural network with stacked LSTM layers significantly improves lipreading accuracy over traditional methods, achieving nearly 80% word recognition on a standard dataset.

Contribution

It introduces an end-to-end neural network architecture with stacked LSTM layers for lipreading, outperforming conventional feature-based classifiers.

Findings

Neural network with LSTM layers achieves 79.6% accuracy.

Outperforms traditional feature-based methods by 11.6%.

Evaluated on the GRID corpus with 19 speakers.

Abstract

Lipreading, i.e. speech recognition from visual-only recordings of a speaker's face, can be achieved with a processing pipeline based solely on neural networks, yielding significantly better accuracy than conventional methods. Feed-forward and recurrent neural network layers (namely Long Short-Term Memory; LSTM) are stacked to form a single structure which is trained by back-propagating error gradients through all the layers. The performance of such a stacked network was experimentally evaluated and compared to a standard Support Vector Machine classifier using conventional computer vision features (Eigenlips and Histograms of Oriented Gradients). The evaluation was performed on data from 19 speakers of the publicly available GRID corpus. With 51 different words to classify, we report a best word accuracy on held-out evaluation speakers of 79.6% using the end-to-end neural network-based solution (11.6% improvement over the best feature-based solution evaluated).