Drop in your exported stems. Get back a folder that makes sense.
StemStash classifies instrument type, detects BPM and key, flags silent takes, and groups hundreds of exported stems into DRUMS, BASS, MELODY, VOCALS, and FX with a MASTER BUS reference, so the files land in your DAW already organized.
Up to 0
files per project
0
analysis stages
0h
storage retention
0
generative calls per stem
Deterministic by design
The same set of stems returns the same classification, BPM, and key on every run. Nothing here is a random guess that changes if you re-upload.
You stay in control
Every detected group, filename, BPM, and key is editable in a review table, with a visible confidence score, before anything exports.
Built for volume
Drag in tens or hundreds of files at once, with resumable uploads and per-file progress.
Interactive preview
See it in action
The same six stems, before and after. Nothing here is uploaded or processed. This is a click-through preview of the real export structure.
Before
- kick (3) FINAL_v2.wav
- New Recording 14.wav
- Untitled-1.wav
- vocal_take_9_use this one.wav
- audio_export_bounce_2am.wav
- Master_v7_ACTUALFINAL.wav
After
Click “Clean up” to preview the organized output.
How the analysis pipeline works
5 stages, per project
Classification
Zero-shot audio classification
Each stem is scored against dozens of instrument and sound categories. Long files are split into consecutive windows and analyzed as a whole, not a single random slice, so the same file returns the same label every time it runs.
Tempo
Onset-based beat tracking, project-wide consensus
Rather than trusting one stem's guess, tempo is built from consensus across every rhythmic file in the project. Drum and bass content is weighted higher, and half-time or double-time reads are folded into a single stable BPM.
Key
Chroma pitch-class correlation
Key detection correlates chroma energy against major and minor key profiles using the Krumhansl-Schmuckler method. Percussion, FX, and silent stems are excluded from the project-level consensus, so a kick drum never skews the result.
Master detection
Evidence-based, never volume alone
The master bus is identified from filename signal, classification confidence, and duration together. The loudest file in a session is never assumed to be the master.
Silence
Peak-based level analysis
Empty takes are flagged from peak level across the full file, not an average. A track with a few sparse hits in an otherwise quiet file is not mistaken for silence.
Output groups
Specification
- Input formats
- WAV, AIFF, FLAC, MP3
- Files per project
- Up to 500
- Max file size
- 50 MB
- Storage retention
- Up to 24 hours after export
- Analysis
- Deterministic, no per-stem generative call