stem-verify
Skill16 starsChanged 36 days agoMITView on GitHub
Verify whether a folder of audio stems sums back to a given master (a "sibling" check) by measuring windowed cancellation depth, correlation, and time offset; triage a folder of audio (levels: dBFS, silence, sample rate, and whether a file is just a copy of a reference); and find where a fragment or take occurs inside a longer reference (locate). Use when confirming exported stems and a master are the same render, identifying which master a stems folder belongs to, spotting silent or duplicate stems, or placing a fragment on the timeline.
In this repository8 other files it gives its agents.
- engine/CLAUDE.mdCLAUDE.md
- als-buildSkill
- als-filesSkill
- als-warpSkill
- engineSkill
- midi-compareSkill
- midi-transcribeSkill
- tempo-driftSkill
The file
Files
You verify that a folder of stems reconstructs a master, using the bundled
Ableton engine (the ableton command on PATH; see the engine skill for the
full command reference and fallback paths).
Invoke
ableton stem-verify --stems <STEMS_DIR> --master <MASTER_FILE> --json
Optional: --win 10 (window seconds), --max-lag-ms 200, --pattern '*.wav'.
Interpret the JSON (you state the verdict — the tool only emits numbers)
worst_db < -15→ true sibling: the stems ARE this master.- median
-30..-10→ similar render / partial match (e.g. different bounce). - median
> -10→ different audio.
Report worst_db, median_db, lag_ms, and pearson_r, then your verdict.
Triage a folder first: levels
ableton levels <DIR> [--ref <FILE>] --json reports each file's duration,
sample rate, dBFS and a silent flag (< −60 dBFS).
With --ref, it also reports r_vs_ref, the lag-0 correlation. Read the
results like this:
r ≈ 1means the file is the reference, e.g. a "vocal bounce" that is actually the backing track.- Silent stems, such as empty guitar or strings separations, can be skipped or
muted on import (
import-stems --skip-below/--mute-below).
Where does a fragment belong? locate
ableton locate --fragment <F> --ref <R> --json returns the best offsets in
seconds plus r, with these bands:
| Band | r | Meaning |
same | ≥ 0.8 | same render; trust the offset |
likely | 0.6–0.8 | probable match |
different | < 0.6 | a different performance |
Independent vocal takes score 0.2–0.6 against each other, so their positions
can't be recovered from audio. Place them by ear, or export them full-length
from the source DAW so the position is baked in.
To test a stems folder against several masters, run once per master and compare
worst_db (most negative wins).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.