Slow-Fast Auditory Streams For Audio Recognition

TL;DR

This paper introduces a two-stream convolutional network for audio recognition that processes spectrograms at different temporal resolutions, achieving state-of-the-art results on VGG-Sound and EPIC-KITCHENS-100 datasets.

Contribution

It presents a novel Slow-Fast two-stream architecture with separable convolutions and lateral connections for improved audio recognition.

Findings

Achieved state-of-the-art accuracy on VGG-Sound dataset.

Achieved state-of-the-art accuracy on EPIC-KITCHENS-100 dataset.

Demonstrated the effectiveness of multi-resolution auditory streams.

Abstract

We propose a two-stream convolutional network for audio recognition, that operates on time-frequency spectrogram inputs. Following similar success in visual recognition, we learn Slow-Fast auditory streams with separable convolutions and multi-level lateral connections. The Slow pathway has high channel capacity while the Fast pathway operates at a fine-grained temporal resolution. We showcase the importance of our two-stream proposal on two diverse datasets: VGG-Sound and EPIC-KITCHENS-100, and achieve state-of-the-art results on both.