Small-Footprint Keyword Spotting on Raw Audio Data with Sinc-Convolutions

TL;DR

This paper introduces an end-to-end keyword spotting model using Sinc-convolutions that directly classifies raw audio, significantly reducing power and memory consumption while maintaining high accuracy.

Contribution

The novel approach eliminates traditional feature extraction, employs Sinc-convolutions for spectral analysis, and uses depthwise separable convolutions to create a compact, efficient model.

Findings

Achieves 96.4% accuracy on Speech Commands dataset.

Uses only 62k parameters, demonstrating high efficiency.

Reduces power and memory consumption compared to previous methods.

Abstract

Keyword Spotting (KWS) enables speech-based user interaction on smart devices. Always-on and battery-powered application scenarios for smart devices put constraints on hardware resources and power consumption, while also demanding high accuracy as well as real-time capability. Previous architectures first extracted acoustic features and then applied a neural network to classify keyword probabilities, optimizing towards memory footprint and execution time. Compared to previous publications, we took additional steps to reduce power and memory consumption without reducing classification accuracy. Power-consuming audio preprocessing and data transfer steps are eliminated by directly classifying from raw audio. For this, our end-to-end architecture extracts spectral features using parametrized Sinc-convolutions. Its memory footprint is further reduced by grouping depthwise separable convolutions. Our network achieves the competitive accuracy of 96.4% on Google's Speech Commands test set with only 62k parameters.