brycewang-stanford INTERSPEECH Submission
Run this audit against the live author pages (Interspeech 2026 kept them under interspeech2026.org → Calls / Author Resources), not against memory. Interspeech re-issues its paper kit every cycle, and the format police here are unusually strict because the whole review pipeline assumes a uniform 4-page object.
The format contract
The regular Interspeech paper is 4 pages of content in two columns, plus one additional page restricted to references and acknowledgments (2025 wording; confirm the 2026 kit). Nothing else goes on page 5. Consequences worth internalizing:
- There is no appendix escape hatch. Whatever reviewers must see lives in the 4 pages.
- Figures compete directly with argument; every plot must earn its column-inches.
- The Long Paper track (introduced for 2026; reported 8 pages + 2 reference pages) has its own rules — do not mix the two templates.
- Use the official paper kit (LaTeX template on the conference site/Overleaf) with fonts, margins, and the PDF checker untouched. Interspeech validates PDFs mechanically at upload.
Anonymity: stricter than you remember
Interspeech runs double-anonymous review, and since 2025 the policy has included an anonymity period that starts one month before the submission deadline and lasts until decisions: no non-anonymized version of the work may be posted online (arXiv included) in that window. Audit for:
- Author names, affiliations, grant numbers, and lab-identifying acknowledgments (acknowledgments generally wait for camera-ready).
- Demo and repository links that resolve to identifiable accounts — speech papers leak identity through audio-sample pages more than through text.
- Self-citation phrasing ("our previous system [3]") and dataset names that only one lab could hold.
- PDF metadata and embedded fonts carrying usernames.
The paper-update window (Interspeech's oddest mechanic)
Interspeech allows the PDF to be revised for a short window after the submission deadline — in 2026, submissions closed 25 February and paper updates were accepted until 4 March. Title, authors, abstract, and area typically freeze at the deadline; only the PDF may change. Plan for it deliberately:
| Days after deadline | Sanctioned use of the update window |
|---|---|
| 0–2 | Full-team proofread of the actually-submitted PDF |
| 2–5 | Fix typos, broken references, figure glitches, overfull boxes |
| 5–close | One final upload; verify the portal shows the new checksum/version |
Do not treat the window as a deadline extension for experiments — reviewers may be bidding on the metadata you froze, and wholesale content swaps invite chair scrutiny.
Metadata and area selection
- The scientific area you pick routes the paper to its reviewer pool; choose for the
claim, not the method (see
interspeech-topic-selection). - Abstract in CMT must match the PDF abstract; it is what reviewers bid on.
- Declare all authors at submission — post-deadline author changes are restricted.
- Answer dual-submission questions honestly: the same work may not sit under review at another conference or journal simultaneously.
Desk-reject triggers observed at this venue
- Fifth page containing anything beyond references/acknowledgments.
- Template tampering (margin, font, or line-spacing tricks) caught by the PDF check.
- Identity leakage anywhere in the PDF or linked material.
- Posting a de-anonymized preprint inside the anonymity period.
- Missing mandatory metadata fields or an invalid PDF at the CMT cutoff.
Final 72 hours, in order
- Rebuild the PDF from a clean clone; confirm the kit's checker passes.
- Anonymity sweep: text, links, metadata, audio-demo pages, repo history.
- Freeze metadata early on CMT — the portal slows near the cutoff.
- Upload, download what CMT stored, and read that file end to end.
- Diary the paper-update close date for the sanctioned typo pass.
CMT field audit, one by one
Portal fields outlive the PDF — the Archive entry, review assignment, and program are generated from them:
- Title: identical to the PDF, punctuation included; it freezes at the deadline in most cycles.
- Abstract: paste from the PDF source, not from an old draft; reviewers bid on this text.
- Authors + order: complete and final; adding an author later needs chair intervention. Verify each co-author's CMT account email — collisions create duplicate identities that break conflict detection.
- Scientific area (+ any secondary): the routing decision from
interspeech-topic-selection; do not delegate it to whoever uploads. - Conflicts/domains: list every co-author's institutional domains; missed conflicts surface as review-integrity problems months later.
- Policy checkboxes (dual submission, ethics, anonymity attestations): read the year's wording — the sentences change, and a false attestation is a desk-reject with memory.
If something goes wrong at the wire
- Portal outage or personal emergency at the cutoff: email the Technical Program Chairs immediately with evidence; discretion exists but shrinks by the hour.
- Wrong file uploaded and the deadline passed: the paper-update window is your sanctioned repair channel for the PDF — but metadata errors need a chair email.
- Discovered identity leak after the deadline: update the PDF within the window and document the fix; do not silently hope.
Output format
[Submission verdict] ready / fixable / not ready
[Format] 4+1 compliance, template integrity, PDF checker result
[Anonymity] text / links / metadata / demo-page findings
[Metadata] area, abstract match, author list, dual-submission answers
[Update-window plan] <what will be checked before the update close>
[Blockers] <ordered list>
Currency note
The 2026 anchor dates — portal open 17 January, deadline 25 February, update close
4 March, all for the Sydney edition — were checked 2026-07-08 via official-page
renderings (see resources/official-source-map.md). Every one of them is
cycle-specific; the 2027 São Paulo cycle will publish its own kit and calendar.