Joint Time-Frequency Scattering for Audio Classification

TL;DR

The paper presents a novel joint time-frequency scattering transform that effectively captures complex audio features, achieving state-of-the-art results in audio classification tasks like phone segmentation.

Contribution

It introduces a new joint time-frequency scattering transform that enhances audio feature representation for classification.

Findings

Successfully characterizes complex time-frequency phenomena

Achieves state-of-the-art results on TIMIT dataset

Effective for signal reconstruction and phone segmentation

Abstract

We introduce the joint time-frequency scattering transform, a time shift invariant descriptor of time-frequency structure for audio classification. It is obtained by applying a two-dimensional wavelet transform in time and log-frequency to a time-frequency wavelet scalogram. We show that this descriptor successfully characterizes complex time-frequency phenomena such as time-varying filters and frequency modulated excitations. State-of-the-art results are achieved for signal reconstruction and phone segment classification on the TIMIT dataset.