ICASSP Supplementary
Use this when assembling everything around an ICASSP paper that does not fit the four content
pages. The defining constraint: ICASSP has no reviewed supplement or appendix that decides
acceptance. The paper must be self-contained in 4+1 pages; anything else is optional material a
reviewer may consult but is not obliged to.
The three shelves
| Shelf |
What lives here |
Reviewer obligation |
| Pages 1-4 |
The claim, mechanism, and decision-critical results and figures |
Must read |
| Page 5 |
References, funding acknowledgements, ethical-standards statement only |
Must read (references) |
| Optional release |
Extended derivations, extra conditions, demo media, code |
May consult, not guaranteed |
If a result decides acceptance, it belongs on pages 1-4. If it merely strengthens or illustrates,
it can live in the optional release — but never assume a reviewer opened it.
What "supplementary" means at ICASSP
- Unlike double-blind ML venues with a reviewed appendix, ICASSP's optional material is usually a
public repository or a demo page, and because review is single-blind it can carry your
identity from the start.
- Some cycles allow uploading a multimedia file (audio, image, or video) with the submission
through CMS; confirm the current paper kit for accepted formats and size limits before relying
on it.
- Long proofs and derivations that will not fit four pages typically go to an extended
arXiv/technical-report version cited from the paper, not to a reviewed appendix.
Curating multimedia reviewers actually open
For enhancement, separation, synthesis, restoration, or any perceptual result, a short set of
listenable or viewable examples is often the most persuasive artifact:
- Provide a small, curated set (roughly five), not a dump; pick cases that show both a strength
and an honest hard case.
- Label each example with its condition (SNR, corpus, or scene) so a reviewer can interpret it.
- For audio, include the input, the output, and a strong-baseline output side by side.
- Scrub embedded metadata only for privacy/consent reasons, not anonymity — single-blind review
does not require hiding your identity.
Keeping the four pages self-contained
- State each main theorem, algorithm, or system so a reviewer can judge it without the external
material; a paper whose argument only closes in the arXiv version reads as unreviewable in four
pages.
- Keep one derivation-sketch line in the body for any result whose full proof you exile to the
technical report.
- Cross-reference the release from the paper at least once; unlinked material is invisible.
Vignette: splitting an image-restoration paper
A submission proposes a denoiser with a stability argument and a broad evaluation. The body keeps
the method, the key stability statement with a one-line justification, and the two
decision-critical PSNR/SSIM tables; the extended derivation and the full noise-level sweep go to a
cited technical report; a demo page shows five before/after crops with a baseline. Nothing that
decides acceptance lives only off-paper, because ICASSP reviewers are not required to leave the
PDF.
Consent and licensing checklist
[On-paper] claim, mechanism, decision-critical results all in 4+1
[Off-paper] derivations / extra conditions / media / code (cited from paper)
[Media] curated set, labeled conditions, baseline shown
[Licensing] corpus/data license permits redistribution of any shipped samples
[Consent] speaker/subject consent cleared for any released recordings or images
Output format
[Supplement plan] ready / needs fixes / not ready
[Page budget] 4 content + 1 reference/acks/ethics respected?
[Self-containment] does the paper close without the external material?
[Media] curated / labeled / baseline-anchored
[Release link] present in paper and reachable
1---2name: icassp-supplementary3description: Use when deciding what goes in the four ICASSP pages versus a public release, given that ICASSP has no reviewed appendix — managing the 4+1 budget, curating optional multimedia (audio, image, video) that reviewers may open, hosting derivations and extra results externally, and keeping the main paper self-contained under single-blind review.4---56# ICASSP Supplementary78Use this when assembling everything around an ICASSP paper that does not fit the four content9pages. The defining constraint: ICASSP has **no reviewed supplement or appendix that decides10acceptance**. The paper must be self-contained in 4+1 pages; anything else is optional material a11reviewer *may* consult but is not obliged to.1213## The three shelves1415| Shelf | What lives here | Reviewer obligation |16|---|---|---|17| Pages 1-4 | The claim, mechanism, and decision-critical results and figures | Must read |18| Page 5 | References, funding acknowledgements, ethical-standards statement only | Must read (references) |19| Optional release | Extended derivations, extra conditions, demo media, code | May consult, not guaranteed |2021If a result decides acceptance, it belongs on pages 1-4. If it merely strengthens or illustrates,22it can live in the optional release — but never assume a reviewer opened it.2324## What "supplementary" means at ICASSP2526- Unlike double-blind ML venues with a reviewed appendix, ICASSP's optional material is usually a27 **public repository or a demo page**, and because review is single-blind it can carry your28 identity from the start.29- Some cycles allow uploading a **multimedia file** (audio, image, or video) with the submission30 through CMS; confirm the current paper kit for accepted formats and size limits before relying31 on it.32- Long proofs and derivations that will not fit four pages typically go to an **extended33 arXiv/technical-report version** cited from the paper, not to a reviewed appendix.3435## Curating multimedia reviewers actually open3637For enhancement, separation, synthesis, restoration, or any perceptual result, a short set of38**listenable or viewable examples** is often the most persuasive artifact:3940- Provide a small, curated set (roughly five), not a dump; pick cases that show both a strength41 and an honest hard case.42- Label each example with its condition (SNR, corpus, or scene) so a reviewer can interpret it.43- For audio, include the input, the output, and a strong-baseline output side by side.44- Scrub embedded metadata only for privacy/consent reasons, not anonymity — single-blind review45 does not require hiding your identity.4647## Keeping the four pages self-contained4849- State each main theorem, algorithm, or system so a reviewer can judge it without the external50 material; a paper whose argument only closes in the arXiv version reads as unreviewable in four51 pages.52- Keep one derivation-sketch line in the body for any result whose full proof you exile to the53 technical report.54- Cross-reference the release from the paper at least once; unlinked material is invisible.5556## Vignette: splitting an image-restoration paper5758A submission proposes a denoiser with a stability argument and a broad evaluation. The body keeps59the method, the key stability statement with a one-line justification, and the two60decision-critical PSNR/SSIM tables; the extended derivation and the full noise-level sweep go to a61cited technical report; a demo page shows five before/after crops with a baseline. Nothing that62decides acceptance lives only off-paper, because ICASSP reviewers are not required to leave the63PDF.6465## Consent and licensing checklist6667```text68[On-paper] claim, mechanism, decision-critical results all in 4+169[Off-paper] derivations / extra conditions / media / code (cited from paper)70[Media] curated set, labeled conditions, baseline shown71[Licensing] corpus/data license permits redistribution of any shipped samples72[Consent] speaker/subject consent cleared for any released recordings or images73```7475## Output format7677```text78[Supplement plan] ready / needs fixes / not ready79[Page budget] 4 content + 1 reference/acks/ethics respected?80[Self-containment] does the paper close without the external material?81[Media] curated / labeled / baseline-anchored82[Release link] present in paper and reachable83```
brycewang-stanford