Video guides

Nine narrated walkthroughs cover TimeCues Studio end to end — from a first tour to writing your own detector and self-hosting the stack. Each is a few minutes. Read the one-line summary, then press play. For a short clip of every single feature, see the full feature walkthrough.

1. Getting started

A three-minute tour of the whole app: signing in, and the five workspaces — Dataprep, Annotator, Algorithm Inspect, Playground, and Team — and where everything lives. Start here.

2. Prepare a dataset

Set a song up for annotation: confirm its tempo with the BPM chips, align the beat grid and bar offset, tap the metronome, and add annotators to the team.

3. Annotate a song

Mark the structure of a track — sections, cues, spans, loops, and lyrics — and see how cues and spans each hold several labelled layers. Covers the annotation visualizer and the keyboard flow.

4. Check the algorithms against your work

Run the built-in algorithms against your own annotations and read the scores — the leaderboard, per-song evaluation, and where each detector agrees or misses.

5. Write your own algorithm

Use the Playground to write a custom structure detector in Python, run it on real audio, and score it alongside the built-in models.

6. Settings — tune the workspace

A tour of every setting: theme, playback, default signals and algorithms, vocabularies and taxonomies, BPM detection, and the reset and danger-zone controls.

7. Self-host on your own server

Run the whole stack on your own machine with Docker Compose — the core services, optional Demucs stems and experimental model sidecars, and the compose profiles that switch them on.

8. Compare two annotators’ work

Load two people’s annotations of the same song side by side and compare where they agree and where they diverge.

9. Separate a song into stems

Split a song into its stems — vocals, drums, bass, and other — with Demucs, and use the isolated tracks while you annotate.