ICASSP Topic Selection

Use this before writing. ICASSP is the IEEE Signal Processing Society flagship, and its defining feature is breadth: speech and audio, image and video, communications and radar, sensor arrays, estimation and detection theory, and machine learning for signals all review under one roof. Speech is one track among many, not the identity of the venue. A project fits when its contribution changes a signal-processing primitive and can be proved in four pages.

Find the signal-processing primitive

Name what the contribution actually is:

• A new estimator, transform, filter, or detector; a new representation or embedding; a new objective or inference procedure; a resource (corpus/benchmark); or a model-based-deep-learning result that changes a block of the signal chain.
• If no signal-processing primitive exists — the paper is generic ML with no signal model — ICASSP is likely the wrong venue (see the re-route table).

The ICASSP-vs-Interspeech routing decision (read this for any speech paper)

Speech papers can go to either venue, and picking wrong costs a cycle. They differ on mechanics as much as on scope:

Decision rule: route to ICASSP when the contribution is a signal-processing method whose speech application is one instance (a new front-end, estimator, separation objective, or array technique), or when the September calendar and single-blind culture fit. Route to Interspeech when the contribution is fundamentally about spoken language — phonetics, prosody, dialogue, speech science, or a speech-specific model where the linguistic content is the point — or when the February calendar and double-blind norms fit. When both fit, let deadline and reviewer community decide.

Fit signal table (full ICASSP breadth)

Re-route table

• Generic ML, no signal model → NeurIPS, ICML, ICLR, AAAI.
• Spoken-language-centric speech → Interspeech (or SLT/ASRU for those subfields).
• Image-specific → ICIP; European/audio-workshop → EUSIPCO, WASPAA, DCASE.
• Journal-length, finished → IEEE TSP, TASLP, SPL, TIP, or a sibling SPS journal; consider the OJSP-ICASSP track to present at ICASSP while publishing open-access in a journal.

Sharpening moves before committing

• State the primitive in one sentence; if you cannot, the ICASSP framing does not yet exist.
• Confirm the result fits four pages with a task-matched metric and a standard baseline; a claim that needs journal-length exposition should go to an SPS journal or the OJSP track.
• For speech work, run the ICASSP-vs-Interspeech table explicitly rather than defaulting to whichever deadline is closer.
• Subject-area emphasis drifts by cycle; scan the current EDICS list before final routing.

Output format

[Fit] strong ICASSP / possible ICASSP / better elsewhere
[Primitive] estimator / transform / representation / objective / resource / model-based-DL / none
[If speech] ICASSP vs Interspeech -> <decision + reason (scope/calendar/blinding)>
[Best venue] ICASSP / Interspeech / ICIP / EUSIPCO / WASPAA / SPS journal / ML venue
[Contribution sentence] <one signal-processing claim>
[Next action] <framing, experiment, or venue switch>