CITATION — REFERENCE ENTRY
The 2018 Signal Separation Evaluation Campaign — Latent Variable Analysis and Signal Separation (LVA/ICA 2018), Lecture Notes in Computer Science
- Key
- stoter2018-sisec
- Authors
- Stöter, Fabian-Robert; Liutkus, Antoine; Ito, Nobutaka
- Issued
- 2018
- Type
- paper-conference
- Container
- Latent Variable Analysis and Signal Separation (LVA/ICA 2018), Lecture Notes in Computer Science
- Publisher
- Springer
- Pages
- 293-305
Raw CSL JSON
{
"DOI": "10.1007/978-3-319-93764-9_28",
"URL": "https://arxiv.org/abs/1804.06267",
"note": "Claims verified from arXiv 1804.06267 v3; page locators refer to that version.",
"page": "293-305",
"type": "paper-conference",
"title": "The 2018 Signal Separation Evaluation Campaign",
"author": [
{
"given": "Fabian-Robert",
"family": "Stöter"
},
{
"given": "Antoine",
"family": "Liutkus"
},
{
"given": "Nobutaka",
"family": "Ito"
}
],
"issued": {
"date-parts": [
[
2018
]
]
},
"publisher": "Springer",
"container-title": "Latent Variable Analysis and Signal Separation (LVA/ICA 2018), Lecture Notes in Computer Science"
}
Claims
-
The widely used BSS Eval metrics assess separation quality by source-to-distortion, source-to-artifact and source-to-interference ratios (SDR, SAR, SIR).
"The BSS Eval metrics, as implemented in the MATLAB toolboxes [7, 32] are widely used in the separation literature. They assess separation quality through 3 criteria: Source to Distortion, to Artefact, to Interference ratios (SDR, SAR, SIR)"
-
The vast majority of methods submitted to SiSEC 2018's music task were based on deep learning, reflecting a shift in the community's methodology.
"the vast majority of methods submitted this year to SiSEC MUS are based on deep learning, reflecting a shift in the community’s methodology."
-
The ideal binary mask, which assigns each time-frequency bin of the mixture to one source only, was used as an oracle method.
"The Ideal Binary Mask (IBM, [35]) is arguably the simplest filtering method. It processes all (f, t, i) of the mixture independently and simply assigns each of them to one source only"
-
The best systems did not perform significantly differently from the oracles for accompaniment, suggesting automatic karaoke is largely solved given enough training data, while vocal separation still had room for improvement.
"they do not perform significantly differently than the oracles for accompaniment, suggesting that the automatic karaoke problem can now be considered solved to a large extent, given sufficient amounts of training data. On the contrary, vocals separation shows room for improvement."
-
MUSDB18 contains 150 full-length tracks, about 10 hours of audio, split into bass, drums, vocals and other, with 100 tracks for training and 50 for testing.
"It contains 150 full-length tracks, totaling approximately 10 h of audio. ... All signals are split into 4 predefined categories: bass, drums, vocals, and other. ... It is split into a training (100 tracks, 6.5 h) and a test set (50 tracks, 3.5 h), for the design of data-driven methods."
Available in