Bayes Risk Transducer: Transducer with Controllable Alignment Prediction

TL;DR

This paper introduces the Bayes Risk Transducer (BRT), a novel ASR transducer that allows controllable alignment prediction by emphasizing preferred paths, leading to significant reductions in inference cost and system latency.

Contribution

The paper proposes the BRT, which incorporates a Bayes risk function to enforce preferred alignments, offering a new way to control transducer predictions in speech recognition.

Findings

BRT reduces inference cost by up to 46% in non-streaming ASR.

BRT decreases system latency by 41% in streaming ASR.

Preferred path enforcement improves practical ASR performance.

Abstract

Automatic speech recognition (ASR) based on transducers is widely used. In training, a transducer maximizes the summed posteriors of all paths. The path with the highest posterior is commonly defined as the predicted alignment between the speech and the transcription. While the vanilla transducer does not have a prior preference for any of the valid paths, this work intends to enforce the preferred paths and achieve controllable alignment prediction. Specifically, this work proposes Bayes Risk Transducer (BRT), which uses a Bayes risk function to set lower risk values to the preferred paths so that the predicted alignment is more likely to satisfy specific desired properties. We further demonstrate that these predicted alignments with intentionally designed properties can provide practical advantages over the vanilla transducer. Experimentally, the proposed BRT saves inference cost by up to 46% for non-streaming ASR and reduces overall system latency by 41% for streaming ASR.