On Training Recurrent Networks with Truncated Backpropagation Through Time in Speech Recognition

TL;DR

This paper investigates how recurrent neural networks learn long-term dependencies in speech recognition, analyzing the effects of decoding strategies and training heuristics on their memory capabilities.

Contribution

It connects decoding approaches with training methods like truncated backpropagation through time, providing insights into the networks' ability to remember long-term dependencies.

Findings

Decoding approach influences the amount of history used for prediction.

Design choices affect the network's ability to capture long-term dependencies.

Connections between Markov processes and vanishing gradients are established.

Abstract

Recurrent neural networks have been the dominant models for many speech and language processing tasks. However, we understand little about the behavior and the class of functions recurrent networks can realize. Moreover, the heuristics used during training complicate the analyses. In this paper, we study recurrent networks' ability to learn long-term dependency in the context of speech recognition. We consider two decoding approaches, online and batch decoding, and show the classes of functions to which the decoding approaches correspond. We then draw a connection between batch decoding and a popular training approach for recurrent networks, truncated backpropagation through time. Changing the decoding approach restricts the amount of past history recurrent networks can use for prediction, allowing us to analyze their ability to remember. Empirically, we utilize long-term dependency in subphonetic states, phonemes, and words, and show how the design decisions, such as the decoding approach, lookahead, context frames, and consecutive prediction, characterize the behavior of recurrent networks. Finally, we draw a connection between Markov processes and vanishing gradients. These results have implications for studying the long-term dependency in speech data and how these properties are learned by recurrent networks.