icassp-related-work

ICASSP Related Work

Use this to audit novelty and positioning. ICASSP is single-blind, so you cite normally — no

anonymized references, no hiding your own prior work. Reopen the current call only for

dual-submission and prior-publication rules. The four-page limit makes related work a

precision instrument, not a survey.

Positioning checks

• Separate the signal-processing novelty (a new estimator, transform, representation,

objective, architecture, or resource) from an engineering improvement.

• Compare against both the IEEE SPS journal literature and the conference literature; a

reviewer from the relevant technical committee knows both.

• Position against the nearest current-cycle work, not a five-year-old baseline; ICASSP

reviewers work in the subfield and notice a stale comparison immediately.

• Because review is single-blind, cite your own prior work in normal first person and make the

incremental delta explicit rather than obscuring it.

Literature lanes table

| Lane | Typical venues | What ICASSP reviewers check |

|---|---|---|

| SPS journals | IEEE TSP, TASLP, SPL, TIP, TIFS, TMM, TCI, TSIPN, JSTSP, OJSP | Whether the closest journal result is acknowledged and the delta stated |

| ICASSP + IEEE conferences | prior ICASSP, ICIP, ICME, SLT, ASRU | Whether the nearest conference method is compared or distinguished |

| Speech/audio siblings | Interspeech, WASPAA, EUSIPCO, DCASE | Whether a paper mis-attributed to ICASSP actually lives here |

| ML venues | NeurIPS, ICML, ICLR, AAAI | Whether generic-ML priors are credited without claiming their generality |

A bibliography that cites only ML venues and no SPS journals tells a committee reviewer the paper

may be rediscovering a known signal-processing result — a recognizable weakness that benchmark

strength does not repair.

The sibling-venue attribution trap

Signal-processing landmarks are scattered across venues, and misattributing one is a credibility

hit. Common confusions to check before citing:

• Papers you think are ICASSP that are actually Interspeech (much of the speech-synthesis

and self-supervised-speech canon) or journal papers (TASLP separation work).

• ICIP for image results and EUSIPCO/WASPAA for European and audio-workshop results.
• Verify each cited venue against dblp (conf/icassp/, journals/) rather than trusting memory;

see [../../resources/exemplars/library.md](../../resources/exemplars/library.md) for the

guard list.

Positioning vignette

Suppose the paper proposes a low-complexity beamformer for a hearing device. Its nearest neighbors:

a TASLP paper with higher complexity, a prior ICASSP paper with a similar structure but no latency

budget, and an ML paper with a black-box network. The novelty sentence should name all three

contrasts — lower complexity than the journal method, an explicit latency budget the prior ICASSP

work lacked, and interpretability the black-box model lacked — in one or two sentences, because

four pages cannot afford a paragraph per contrast.

Concurrent-work judgment

• Independently concurrent arXiv work: cite neutrally, state the technical difference, and avoid

priority claims a reviewer cannot verify.

• Your own workshop or prior-conference version: cite it in first person and state exactly what

this paper adds, since single-blind review does not require hiding it.

• If a venue's archival status is unclear for dual-submission purposes, declare the overlap rather

than gambling on an organizer's interpretation.

Output format

[Novelty type] estimator / transform / representation / objective / architecture / resource
[Nearest 3 works] <work -> venue (verified) -> technical delta>
[Lane coverage] SPS journals / ICASSP+conf / siblings / ML — gaps?
[Attribution risk] <any venue to re-verify on dblp>
[Novelty sentence] <one ICASSP-ready contribution contrast>