Convolutional REpresentation for Pitch Estimation

Status:
Estimated Pitch:
Voicing Confidence:


NOTE: This demo works most reliably in the latest versions of Chrome on Windows and MacOS. There is an issue where TensorFlow.js is not working properly on Linux versions of Chrome.

A stripped-down model of CREPE is running on this browser, which has less than 3 percent of parameters. The performance of this online demo therefore may not be as good as reported in the paper, and it may make more octave errors than the full model. To run the model with the full capacity, check out the Git repo at https://github.com/marl/crepe and follow the instructions in README.md.

The model is trained on 16 kHz audio, and due to the imperfect resampling in the browser, this demo works best when the hardware sample rate is a multiple of 16000 Hz. Your sample rate is Hz.

For further details on this model, please refer to our paper:

    CREPE: A Convolutional Representation for Pitch Estimation
    Jong Wook Kim, Justin Salamon, Peter Li, Juan Pablo Bello
    Proceedings of the IEEE Conference on ICASSP, 2018.