Incorporation of phase information for improved time-dependent instrument recognition
Time-dependent estimation of playing instruments in music recordings is an important preprocessing for several music signal processing algorithms. In this approach, instrument recognition is realized by neural networks with a two-dimensional input of short-time Fourier transform (STFT) magnitudes an...
Saved in:
Published in | Technisches Messen Vol. 87; no. 1; pp. 62 - 67 |
---|---|
Main Authors | , , |
Format | Journal Article |
Language | English |
Published |
De Gruyter
25.09.2020
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Time-dependent estimation of playing instruments in music recordings is an important preprocessing for several music signal processing algorithms. In this approach, instrument recognition is realized by neural networks with a two-dimensional input of short-time Fourier transform (STFT) magnitudes and a time-frequency representation based on phase information. The modified group delay (MODGD) function and the product spectrum (PS), which is based on MODGD, are analysed as phase representations. Training and evaluation processes are executed based on the MusicNet dataset. By the incorporation of PS in the input, instrument recognition can be improved about 2% in F1-score. |
---|---|
ISSN: | 0171-8096 2196-7113 |
DOI: | 10.1515/teme-2020-0031 |