Denoise data of different sampling frequency

[jnjnnjzch]

Hi,

I'm dealing with my recorded data with dartsort, but the errors raise as the sampling fequency become 20 kHz instead of the default 30 kHz.

I've noticed that in your paper of dartsort and YASS the denoiser is implemented with a neural network, and in code it is implemented with a pretrained NN. So I'm wondering if this provided denoiser NN is suitble for different sampling frequency data?

Besides, it mentioned another pipeline that avoid using NN to denoise data. May I know how to use it with your codes?

Looking for any response, thanks!

[cwindolf]

Hi @jnjnnjzch , thanks for reaching out!

You can turn off the NN by setting do_nn_denoise=False in any FeaturizationConfigs that appear in your pipeline. Since the iterative subtraction in our initial detection works best with the NN, you'll also want to set detection_thresholds=[5] (or your threshold of choice) to avoid the iterations in your SubtractionConfig. Note that configuration will become simpler as we get closer to having a spike sorting pipeline that is ready for users.

Given that this code is currently experimental, please be aware that there will be breaking changes in the APIs and that there may be issues which appear as we continue to work on this code. I'd be curious to learn more about your usage of dartsort so far! Feel free to email me if you like (email in my profile).