Improving RNN Transducer With Target Speaker Extraction and Neural Uncertainty Estimation

TL;DR

This paper introduces a joint framework combining target-speaker speech extraction with RNN-T, utilizing neural uncertainty estimation to improve speech recognition accuracy in noisy multi-speaker environments.

Contribution

It proposes a multi-stage training strategy and neural uncertainty measures to enhance RNN-T performance in challenging acoustic conditions.

Findings

Achieves 17% relative CER reduction with neural uncertainty module.

Gains 9% relative performance improvement in noisy environments.

Maintains performance in clean conditions.

Abstract

Target-speaker speech recognition aims to recognize target-speaker speech from noisy environments with background noise and interfering speakers. This work presents a joint framework that combines time-domain target-speaker speech extraction and Recurrent Neural Network Transducer (RNN-T). To stabilize the joint-training, we propose a multi-stage training strategy that pre-trains and fine-tunes each module in the system before joint-training. Meanwhile, speaker identity and speech enhancement uncertainty measures are proposed to compensate for residual noise and artifacts from the target speech extraction module. Compared to a recognizer fine-tuned with a target speech extraction model, our experiments show that adding the neural uncertainty module significantly reduces 17% relative Character Error Rate (CER) on multi-speaker signals with background noise. The multi-condition experiments indicate that our method can achieve 9% relative performance gain in the noisy condition while maintaining the performance in the clean condition.