Semantic VAD: Low-Latency Voice Activity Detection for Speech Interaction

TL;DR

This paper introduces a semantic VAD method that significantly reduces latency in speech segmentation by incorporating frame-level punctuation prediction and semantic loss, improving user experience in speech interactions.

Contribution

It proposes a novel low-latency semantic VAD with frame-level punctuation prediction and semantic loss, outperforming traditional VAD in latency reduction.

Findings

Latency reduced by 53.3%

No significant increase in character error rate

Effective in internal dataset evaluations

Abstract

For speech interaction, voice activity detection (VAD) is often used as a front-end. However, traditional VAD algorithms usually need to wait for a continuous tail silence to reach a preset maximum duration before segmentation, resulting in a large latency that affects user experience. In this paper, we propose a novel semantic VAD for low-latency segmentation. Different from existing methods, a frame-level punctuation prediction task is added to the semantic VAD, and the artificial endpoint is included in the classification category in addition to the often-used speech presence and absence. To enhance the semantic information of the model, we also incorporate an automatic speech recognition (ASR) related semantic loss. Evaluations on an internal dataset show that the proposed method can reduce the average latency by 53.3% without significant deterioration of character error rate in the back-end ASR compared to the traditional VAD approach.