How to discard the first N seconds of the neural data before sorting?

[georginjacob]

Hi,
I am setting up spikeinterface for the neural data sampled at 25000 kHz from 256 channels. I have found that there is a ramp period of 5 seconds for all the channels in the raw data. While running the sorter on the full data, I am finding a lot of spikes in this initial 5 seconds. Kindly suggest a way to discard the initial 5 second data before passing the data into the sorter or make all the values of this initial 5 seconds as zeros. We are very sure that our experiment starts much later that this time.

The below figures shows the detected spikes from all 256 channels in the first 20 seconds.

Note: I have tried changing the "file_offset" parameter of si.read_binary to match required samples to skip (5 x 25000 x 2 x 256 : seconds x sampling freq x bytes per sample x channels), reading binary data works but ss.run_sorter api doesn't work.

Thanks in advace

Regards,
Georgin

[JoeZiminski]

Hi Georgin,

I think one option is the silence periods preprocessing tool. This can be used to set all samples within a region (expresssed in indicies) to zero. In your case, to remove the first 5 seconds (assuming you have only 1 segment) the input to the list_periods argument I believe should be [(0, 125000)] (assuming a sampling rate of 25 kHz or 25000 Hz rather than 25000 kHz).

[georginjacob]

Hi,

Thanks for the pointer to silence period api. The suggested solution worked. Sorry for the typo, sampling period is 25,000 Hz.

Plots below shows the raw signal and generated spikes after applying silencing.

Regards,
Georgin

[alejoe91]

Thanks @JoeZiminski

If the artifact is always at the beginning, another option is just to clip the signal to start at 5s. You can use the frame_slice function:

recording_cut = recording.frame_slice(start_frame=int(5*recording.sampling_frequency), end_frame=recording.get_num_samples())

The difference is that with this approach you will not have 5 seconds of blank zeros.