Stacked 1D convolutional networks for end-to-end small footprint voice trigger detection

TL;DR

This paper introduces a stacked 1D convolutional neural network (S1DCNN) for efficient, end-to-end voice trigger detection on small devices, outperforming SVDF-based models in reducing false rejects while maintaining low resource usage.

Contribution

The paper presents S1DCNN, a novel model architecture that generalizes SVDFs and improves false reject rates in small-footprint voice trigger detection systems.

Findings

S1DCNN achieves 19% relative FRR reduction over SVDF.

S1DCNN maintains similar model size and latency as SVDF.

Longer time delays further improve FRR by up to 12.2%.

Abstract

We propose a stacked 1D convolutional neural network (S1DCNN) for end-to-end small footprint voice trigger detection in a streaming scenario. Voice trigger detection is an important speech application, with which users can activate their devices by simply saying a keyword or phrase. Due to privacy and latency reasons, a voice trigger detection system should run on an always-on processor on device. Therefore, having small memory and compute cost is crucial for a voice trigger detection system. Recently, singular value decomposition filters (SVDFs) has been used for end-to-end voice trigger detection. The SVDFs approximate a fully-connected layer with a low rank approximation, which reduces the number of model parameters. In this work, we propose S1DCNN as an alternative approach for end-to-end small-footprint voice trigger detection. An S1DCNN layer consists of a 1D convolution layer followed by a depth-wise 1D convolution layer. We show that the SVDF can be expressed as a special case of the S1DCNN layer. Experimental results show that the S1DCNN achieve 19.0% relative false reject ratio (FRR) reduction with a similar model size and a similar time delay compared to the SVDF. By using longer time delays, the S1DCNN further improve the FRR up to 12.2% relative.