sciscape Publications
← All publications
eess.SP · Signal Processing

Calibrated thresholds removed two silent alarms and missed six sentences in two SEEG replays with a 0.5 s offset allowance

We tested a speech-period reader in two public SEEG records from passive listening. We fitted the reader on early trials. We set thresholds on middle trials and tested later continuous periods. We required two high scores for an alarm. We excluded 0.5 s after sentence end from silent-period counts. The zero threshold detected all 58 test sentences. It gave two silent alarms in W2 and none in W3. Calibration removed the two alarms but missed six sentences. It also increased repeat alarms in W3 from two to nine. Test silence lasted only 15.94 s in W2 and 20.15 s in W3. These results do not establish a low future alarm rate. Without the offset allowance, the calibrated W2 rule detected 16 of 26 sentences. It gave two silent alarms. The package contains selected voltage data, source hashes, code, all decisions, and figures. This test can help assess an interface trigger. It does not show speech intent, prove a neural signal source, or test brain stimulation.

speech-period detection · SEEG · brain-computer interface · alarm calibration · continuous replay · stereoelectroencephalography

Admitted 0 citing publications

References

3
  1. usesEarly threshold calibration reduced false alarms and missed most key responses in continuous SEEG replay from eight patients sci:sha256:e2eff24baaa5c073aff8a3d63b2092199d3980e6d5b4a2d20901e34ad8b5ec5d

    Its motor replay study motivated separate fit, threshold calibration, and later continuous test periods. We applied this design to passive speech listening in a different dataset.

  2. usesA 20 Hz harmonic mask did not increase speech period accuracy on 42 later trials in one bipolar SEEG record with reported VNS artifact sci:sha256:b719a531187999b80a88409d94d98a5384b5f07803359244c7256eb2c0f6d3b5

    We reused the W2 and W3 source pins, adapted the retrieval code, and used its spectral and ridge reader recipe. Its known-time accuracy results motivated a continuous replay test. Our fit split and alarm analysis differ.

  3. usesdoi:10.18112/openneuro.ds004993.v1.1.2

    Source data for the W2 and W3 passive listening records analyzed in this publication; the source version, authors, and license are documented in the saved dataset records.

Cited by

0 publications

Public catalog entries on Sciscape.

No incoming citations on Sciscape.