Multichannel Voice Trigger Detection Based on Transform-average-concatenate
Takuya Higuchi, Avamarie Brueggeman, Masood Delfarah, Stephen Shum
TL;DR
This work tackles voice trigger detection by leveraging multichannel outputs from a front-end system, addressing information loss in channel selection. It introduces a Transform-average-concatenate (TAC) block-based multichannel VT model and a modified TAC that conditions on the conventionally selected channel, enabling attention to the target speaker in mixtures. Through extensive on-device–oriented design, including self-attention pooling to reduce cost, the approach achieves up to a 30% relative FRR reduction in quiet and notable improvements in playback scenarios, demonstrating the practicality of multichannel VT for robust wake-word detection. The results suggest that integrating multichannel information with channel-selection cues provides a scalable, efficient path toward more reliable on-device VT in real-world environments.
Abstract
Voice triggering (VT) enables users to activate their devices by just speaking a trigger phrase. A front-end system is typically used to perform speech enhancement and/or separation, and produces multiple enhanced and/or separated signals. Since conventional VT systems take only single-channel audio as input, channel selection is performed. A drawback of this approach is that unselected channels are discarded, even if the discarded channels could contain useful information for VT. In this work, we propose multichannel acoustic models for VT, where the multichannel output from the frond-end is fed directly into a VT model. We adopt a transform-average-concatenate (TAC) block and modify the TAC block by incorporating the channel from the conventional channel selection so that the model can attend to a target speaker when multiple speakers are present. The proposed approach achieves up to 30% reduction in the false rejection rate compared to the baseline channel selection approach.