2022
DOI: 10.48550/arxiv.2202.07968
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

On loss functions and evaluation metrics for music source separation

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1

Citation Types

1
1
0

Publication Types

Select...
1

Relationship

0
1

Authors

Journals

citations
Cited by 1 publication
(2 citation statements)
references
References 0 publications
1
1
0
Order By: Relevance
“…On the other hand, the time-frequency losses showed us a promising result for the L3DAS22 dataset, especially the L 1−f req loss, which is a consistent result with the speech enhancement and source separation literature (GUSÓ et al, 2022;NAGANO;SILVA, 2020;PANDEY;WANG, 2018). Observing our three loss functions, we can see that the STOI metric has a different but not significant as the WER metric.…”
Section: Supplementary Studies On the L3das22 Datasetsupporting
confidence: 71%
See 1 more Smart Citation
“…On the other hand, the time-frequency losses showed us a promising result for the L3DAS22 dataset, especially the L 1−f req loss, which is a consistent result with the speech enhancement and source separation literature (GUSÓ et al, 2022;NAGANO;SILVA, 2020;PANDEY;WANG, 2018). Observing our three loss functions, we can see that the STOI metric has a different but not significant as the WER metric.…”
Section: Supplementary Studies On the L3das22 Datasetsupporting
confidence: 71%
“…Most neural networks are optimized with the Gradient Descent method, so our loss function must be differentiable. Similar to the work of (GUSÓ et al, 2022), who studied the impacts of loss functions on the problem of music source separation, we further investigate losses for speech enhancement.…”
Section: On Loss Functions For Speech Enhancementmentioning
confidence: 99%