onnxcrepe

ONNX deployment of the CREPE [1] pitch tracker. The provided model weights and most of the codes in this repository were converted and migrated from the original TensorFlow implementation here and Max Morrison's torchcrepe, a PyTorch implementation of CREPE.

Usage

Download model weights from releases and put them into the onnxcrepe/assets/ directory. See demo here.

Documentation of this repository is still a work in progress and is comming soon.

Acknowledgements

Codes and model weights in this repository are based on the following repos:

• torchcrepe for 'full' and 'tiny' model weights and most of the code implementation
• Weights_Keras_2_Pytorch for converting 'large', 'medium' and 'small' model weights from the original implementation
• PyTorch for exporting onnx models
• onnx-optimizer and onnx-simplifier for optimizing performance
• onnxruntime for execution and configurations

References

[1] J. W. Kim, J. Salamon, P. Li, and J. P. Bello, “Crepe: A Convolutional Representation for Pitch Estimation,” in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).