Deliberation of Streaming RNN-Transducer by Non-autoregressive Decoding

TL;DR

This paper introduces a non-autoregressive decoding method for streaming RNN-T models that refines hypothesis alignments through multiple steps, improving recognition accuracy with minimal additional parameters.

Contribution

It proposes a novel alignment refinement approach using a transformer decoder with CTC training, enhancing streaming RNN-T performance.

Findings

Significantly improved recognition accuracy over first-pass RNN-T.

Effective alignment refinement with minimal increase in model complexity.

Enhanced audio context capture through cascaded encoder and alignment augmentation.

Abstract

We propose to deliberate the hypothesis alignment of a streaming RNN-T model with the previously proposed Align-Refine non-autoregressive decoding method and its improved versions. The method performs a few refinement steps, where each step shares a transformer decoder that attends to both text features (extracted from alignments) and audio features, and outputs complete updated alignments. The transformer decoder is trained with the CTC loss which facilitates parallel greedy decoding, and performs full-context attention to capture label dependencies. We improve Align-Refine by introducing cascaded encoder that captures more audio context before refinement, and alignment augmentation which enforces learning label dependency. We show that, conditioned on hypothesis alignments of a streaming RNN-T model, our method obtains significantly more accurate recognition results than the first-pass RNN-T, with only small amount of model parameters.