ipt_models

Pretrained instrumental playing-technique recognition models for ipt~, pipo.ipt and any host built on libipt.

Models are TorchScript (.ts) files trained and exported with ipt_recognition. Documentation, classes and specifications of each model live in the ipt_models GitHub repository.

Model	Instrument	Classes	Sample rate	Latency floor	Dataset	Card
eguitar_ircam.ts	Electric guitar	14	8 kHz	896 ms	EG-IPT	card
flute_ircam.ts	Flute	11	44.1 kHz	333 ms	GFD	card
trumpet_ircam.ts	Trumpet	14	44.1 kHz	333 ms	T-IPT (not public yet)	card
trumpet_harmon_ircam.ts	Trumpet, harmon mute	14	44.1 kHz	333 ms	T-IPT (not public yet)	card

Usage

Download a .ts file, put it in Max's search path (or give an absolute path), then:

ipt~ flute_ircam.ts
pipo~ ipt @ipt.model flute_ircam.ts

Or from the command line:

hf download nbrochec/ipt_models flute_ircam.ts --local-dir .

A model takes (batch, 1, segment_length) float32 raw audio and returns logits (batch, num_classes); it exports get_sr(), get_seglen() and get_classnames(). Exported with torch 2.4.1, the version libipt links against.

License

CC-BY-NC-4.0.

Contact

Please write to nicolas.brochec[at]ircam.fr for any questions.