Cross-species / Annotation software

DAS: build on an existing annotation workflow

Source year 2021–2026 · Note reviewed 2026-10-01

Deep Audio Segmenter connects manual labels, model training, automatic proposals and human correction in an existing audio workflow.

What we reviewed · Documentation reviewed

Official introduction, repository README and bird tutorial through proofreading. The primary performance paper and current package were not evaluated.

Not yet independently reviewed by a scientific specialist.

What this does — and doesn’t — establish

Suggested labels still need checking. A quick-start training split is not a validation protocol for our intended species or new recording conditions.

What we can build on

Keep model suggestions separate from accepted observations. Test file interoperability and reviewer effort before building another full annotation editor.

A practical starting point

Candidate for interoperability

Available starting point: DAS graphical annotation and proposal-review workflow

Reuse limits: Repository declares Apache-2.0 for code. Example audio, trained models and dependencies need separate checks.

Our proposed experiment: Exchange a synthetic WAV and interval labels, then compare whether event boundaries and labels survive the round trip.

The next reading task

Inspect the exact DAS import/export schema and run a synthetic round trip before proposing an adapter.

Go to the sources

Deep Audio Segmenter documentation ↗
Official DAS documentation for annotation, training and prediction. Documentation review; not an installation test.
Deep Audio Segmenter code ↗
Official janclemenslab repository; README and Apache-2.0 code-license declaration reviewed.
DAS bird annotation tutorial ↗
Official bird tutorial: annotation, training, prediction and human proofreading workflow reviewed.

Original editorial note by Talk2Nature contributors. AI-assisted, source-grounded synthesis; not a peer review or an independent replication. Linked works retain their own rights.