Real-time inference workshop models
The ONNX models of the realtime-inference-workshop,
downloaded at configure time by its cmake/models.cmake.
rave/model_forward.onnx audio + state in, audio + state out
rave/model_encoder.onnx audio + state in, 8 latents + state out
rave/model_decoder.onnx 8 latents + state in, audio + state out
handtracking/ MediaPipe's four hand tracking models, for step 11
The rave models take 2048 samples (or one latent frame) per call, 44.1 kHz, one channel.
Exported with scripts/export_model.py from the Intelligent Instruments Lab's
magnetic resonator piano model mrp_strengjavera_b2048_r44100_z16.ts
(Intelligent-Instruments-Lab/rave-models, CC BY-NC 4.0).
The hand tracking models in handtracking/ are Google's MediaPipe gesture recognizer as ONNX
(Apache 2.0), the same files as in
anira-project/mediapipe-gesture-recognizer.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support