ipt~ is a Max/MSP external object for real-time classification of instrumental playing techniques.
This object loads and runs TorchScript (.ts) classification models, enabling low latency inference on CPU and MPS devices. The package also includes pipo.ipt, the official PiPo module, which brings the same recognition to MuBu, both in real time and offline.
pipo.iptis the offical PiPo module for MuBu processing; batch mode is used automatically for offline hosts such asmubu.process, developed with diemoschwarz.- Inference is now powered by libipt, a standalone C library.
- New attributes:
@periodallows you to throttle the output of ipt~ while keeping leaky integrator smoothing. - Addition of two trumpet models (with and without harmon mute) to the ipt~ Max Package.
- macOS 10.13 or later with an Apple Silicon processor (M1 or later — Intel Macs are not supported), or Windows 10/11 on an x64 CPU with AVX2 (any desktop CPU from roughly 2013 onward)
- Max 8.6 or later / Max 9.0.3 or later
- Go to Releases and download the latest version of ipt~
- Run the installer depending on your version of Max and follow the instructions
- ipt_recognition: train your own playing techniques recognition model
- ipt_models: pretrained models (electric guitar, flute, trumpet) and their cards, downloadable from Hugging Face
- libipt: the standalone C library behind ipt~, to use it in your own project
- IPT VAMP Plugin by pierreguillot
This project is part of an ongoing research effort into the real-time recognition of instrumental playing techniques for interactive music systems.
If you use this work in your paper, please cite the references listed in CITATION.md, where you will also find our related publications.
This project is released under a CC-BY-NC-4.0 license.
This research is supported by the European Research Council (ERC) as part of the Raising Co-creativity in Cyber-Human Musicianship (REACH) Project directed by Gérard Assayag, under the European Union's Horizon 2020 research and innovation program (GA #883313). Funding support for this work was provided by a Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) scholarship to Nicolas Brochec.
Please write to nicolas.brochec[at]ircam.fr for any questions, or to share with us project made with ipt~
Building from source: BUILDING.md. Repository internals, threading rules and conventions (for developers, and for LLM agents): AGENTS.md