The open‑source plugin NeuralNote adds state‑of‑the‑art audio‑to‑MIDI conversion to any Digital Audio Workstation. It works with vocal and instrumental sources, supports polyphonic transcription, and detects pitch bends. Installers are provided for Windows, macOS (Universal) and Linux, offering VST3, AU (Mac only) and standalone formats.

NeuralNote runs Spotify's basic-pitch model via RTNeural for the convolutional network and ONNXRuntime for feature extraction. The pipeline uses a Constant‑Q transform and harmonic stacking, delivering fast, lightweight transcription that can be tweaked in real time. Users can drop audio files or record directly, then drag the generated MIDI into a track.

The project ships under Apache‑2.0, includes source for the transcription engine, and invites contributions. While not suited for live performance due to CQT latency and a 120 ms CNN delay, it offers a practical workflow for producers needing quick, accurate MIDI extracts. The codebase also adds 2D convolution support to RTNeural, expanding its utility for other audio AI tools.