Showing 1–1 of 1 results for author: Muth, J
-
Improving DNN-based Music Source Separation using Phase Features
Authors:
Joachim Muth,
Stefan Uhlich,
Nathanael Perraudin,
Thomas Kemp,
Fabien Cardinaux,
Yuki Mitsufuji
Abstract:
Music source separation with deep neural networks typically relies only on amplitude features. In this paper we show that additional phase features can improve the separation performance. Using the theoretical relationship between STFT phase and amplitude, we conjecture that derivatives of the phase are a good feature representation opposed to the raw phase. We verify this conjecture experimentall…
▽ More
Music source separation with deep neural networks typically relies only on amplitude features. In this paper we show that additional phase features can improve the separation performance. Using the theoretical relationship between STFT phase and amplitude, we conjecture that derivatives of the phase are a good feature representation opposed to the raw phase. We verify this conjecture experimentally and propose a new DNN architecture which combines amplitude and phase. This joint approach achieves a better signal-to distortion ratio on the DSD100 dataset for all instruments compared to a network that uses only amplitude features. Especially, the bass instrument benefits from the phase information.
△ Less
Submitted 16 July, 2018; v1 submitted 7 July, 2018;
originally announced July 2018.
