Joint Time-Frequency Scattering for Audio Classification
This work addresses audio classification challenges, particularly for tasks like phone recognition, but appears incremental as it builds on existing wavelet-based methods.
The paper tackled the problem of audio classification by introducing the joint time-frequency scattering transform, a time shift invariant descriptor, and achieved state-of-the-art results for signal reconstruction and phone segment classification on the TIMIT dataset.
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.