Tools for processing stim data from Sam's DynamicRoutingTask, compatible with data in the cloud.
conda create -n npc_samstim python>=3.11
conda activate npc_samstim
pip install npc_samstim>>> import npc_samstimSessions recorded before 2023-08-31 have no "sound running" line on sync, so sound
onsets are found in the NI-DAQ recording of the speaker signal with
get_stim_latencies_from_nidaq_recording(..., waveform_type='sound').
The default method (waveforms.align_sound, core in sound_alignment.py):
- models playback (NI-DAQ AO zero-order hold at 80 MHz/1667 = 47990.4 Hz, or a sound card's band-limited output) and calibrates the effective rate per stim file,
- finds each onset with a robust envelope correlation, refined by carrier correlation for noise,
- maps NI-DAQ samples to sync time through all matched barcodes (not just the first and last),
- returns NaN latency when the recording doesn't match the waveform.
Validated against the sync line in 33 sessions (21,069 trials):
- AM-noise onsets: p95 error ~10 us
- tone onsets: p95 ~0.12 ms, and at most 0.29 ms for any single frequency
The previous raw cross-correlation (xcorr_legacy, still used for opto, or via legacy=True)
was more than 1 ms wrong for 85% of tone trials, by up to ~100 ms.
See outputs/aud-onset-xcorr/report.pdf and experiments/aud_xcorr/.
See instructions in https://github.com/AllenInstitute/npc_samstim/CONTRIBUTING.md and the original template: https://github.com/AllenInstitute/copier-pdm-npc/blob/main/README.md