Skip to content
← Models
Open weightsUniversity of Rochester (ISMIR 2021 research)/10 trustVerified

BeatNet

Real-time and offline joint beat, downbeat, tempo, and meter tracking.

Visit official page →

About

BeatNet pairs a convolutional-recurrent neural network, which emits beat and downbeat activations per audio frame, with a two-stage particle filter that estimates tempo, beats, downbeats, and musical meter. The ISMIR 2021 paper reported that the design beats prior online and offline trackers, and the information-gate trick in the particle filter keeps the CPU cost down: on a desktop Ryzen 9, a 30-second excerpt processes in a few seconds.

The library ships with pre-trained models, four modes (streaming from a microphone, real-time, online, and offline), and since v1.2.0 the official training pipeline is included so you can train the CRNN from scratch. It is a PyPI package (pip install BeatNet) under CC-BY-4.0. Worth knowing: newer transformer-based trackers like Beat This! (ISMIR 2024, MIT) have since taken the accuracy crown in the eyes of some practitioners, so BeatNet is best viewed as the solid, well-documented incumbent rather than the frontier.

Specs

Context window

Audio input (streaming or file-based)

Pricing

Free — open source (CC-BY-4.0)

Free quota

N/A (local inference)

API access

No hosted API — local Python library with bundled pre-trained models

Weights

Open weights

Released

Not disclosed

Benchmarks

Provider-reported: the ISMIR 2021 paper compares BeatNet against online and offline baselines and reports superior beat and downbeat tracking; numbers are provider-claimed and independent replication is limited.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

beat-trackingtempo-detectiondownbeat-trackingdj-toolsmusic-analysis

Trust score — 7/10

/10 trustVerified

Scored 7/10 (verified, metrics-v1). Evidence: ISMIR 2021 paper (arXiv 2108.03576); original repo mjhydri/BeatNet with ~524 stars and CC-BY-4.0 license; actively maintained fork another-blue/beatnet added the official training pipeline (v1.2.0). Strengths: pre-trained models bundled, real-time capable, permissive license. Docked for: (1) beat/downbeat accuracy claims are provider-reported; (2) newer transformer trackers (e.g. Beat This!) have superseded it for some users. License: CC-BY-4.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
6.0/10
Maintenance · recency, releases, issue load
6.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · ISMIR 2021 paper (arXiv 2108.03576): CRNN + two-stage particle filter for joint beat/downbeat/tempo/meter
  • · Original repo mjhydri/BeatNet (~524 stars); community fork another-blue/beatnet added the official training pipeline (v1.2.0)
  • · Accuracy claims are provider-reported; newer transformer trackers (e.g. Beat This!, ISMIR 2024) have superseded it for some users

Sources