You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

AudioEstimate Model

AudioEstimate checkpoint for local music-library key and tempo estimation.

Model Description

This model estimates musical key in Camelot notation and BPM/tempo from audio. It is used by the AudioEstimate Python package and Electron desktop app.

The model has two main branches:

key branch: predicts 24 Camelot key classes
tempo branch: predicts BPM using DeepRhythm-compatible HCQM tempo features

The app also reports confidence values so users can review lower-confidence predictions before writing tags or filenames.

Intended Use

Intended for private/local DJ-library analysis:

estimate Camelot key
estimate BPM
review predictions and confidence
optionally write title/key/BPM metadata and/or rename files

This is not intended as authoritative musicological ground truth.

Files

model.pt: PyTorch checkpoint
config.yaml: matching inference/training config
metrics.json: private summary of current behavior
VERSION.md: current artifact note

Inference Convention

Current project convention:

normalize low BPM predictions below 90 BPM by doubling
keep raw BPM as diagnostic metadata
keep key and tempo confidence for review/fallback workflows

Known Limitations

Tempo octave ambiguity can still occur.
Triplet/swing material can be difficult and may be labeled differently by other tools.
Sparse/no-drum sections can reduce tempo reliability.
Accuracy depends on how close the target music library is to the training distribution.
Key predictions can confuse musically related/adjacent Camelot classes.

Downloads last month: 4