Prediction-Adaptation-Correction Recurrent Neural Networks for Low-Resource Language Speech Recognition

TL;DR

This paper introduces PAC-RNNs, a novel neural network architecture for low-resource speech recognition, leveraging prediction and correction modules with transfer learning to outperform existing models.

Contribution

The paper proposes PAC-RNNs, combining prediction and correction networks with transfer learning, achieving superior performance in low-resource speech recognition tasks.

Findings

PAC-RNNs outperform DNNs and LSTMs on IARPA-Babel tasks.

Transfer learning from similar languages improves recognition accuracy.

The model effectively utilizes auxiliary information for better state estimation.

Abstract

In this paper, we investigate the use of prediction-adaptation-correction recurrent neural networks (PAC-RNNs) for low-resource speech recognition. A PAC-RNN is comprised of a pair of neural networks in which a {\it correction} network uses auxiliary information given by a {\it prediction} network to help estimate the state probability. The information from the correction network is also used by the prediction network in a recurrent loop. Our model outperforms other state-of-the-art neural networks (DNNs, LSTMs) on IARPA-Babel tasks. Moreover, transfer learning from a language that is similar to the target language can help improve performance further.