Improving RNN Transducer With Target Speaker Extraction and Neural Uncertainty Estimation
This work provides an incremental improvement for speech recognition systems operating in noisy, multi-speaker environments, benefiting users who need accurate transcription in challenging acoustic conditions.
This paper addresses target-speaker speech recognition in noisy environments by integrating time-domain target-speaker speech extraction with a Recurrent Neural Network Transducer (RNN-T). The proposed method, which includes neural uncertainty estimation, achieves a 17% relative Character Error Rate (CER) reduction on multi-speaker signals with background noise and a 9% relative performance gain in noisy conditions.
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.