sciscape Publications
← All publications
eess.AS · Audio and Speech Processingq-bio.NC · Neurons and Cognition

Within-speech spectral shape exceeded a shift null in five SEEG participants, while spectrogram r matched envelope-only r

This replication and extension adds participant sub-05 from SingleWordProductionDutch-iBIDS to earlier SCI analyses, enabling the preregistered group test at n = 5. Using unchanged offline linear high-gamma decoders on pitch-shifted Dutch single-word audio, the standard mean-bin spectrogram correlation (M0) matched the envelope-only correlation (M1) across 10 participant-decoder cases (largest absolute difference 0.0016; median M1/M0 0.9999 for SYM and 0.9997 for PAST). Within-speech spectral-shape correlation (M3) exceeded the shift null in all five participants. The one-sided Wilcoxon signed-rank p was 0.0312 for both symmetric (SYM) and causal (PAST) decoders, the smallest possible value at n = 5. The median M3 was 0.145 for SYM and 0.146 for PAST; the median null values were 0.027 and 0.029. Mean decoded spectral shape did not identify words above chance, while envelope time course did. The group includes five of ten dataset participants; sub-06 and sub-08 analyses did not finish, sub-07 feature extraction failed, and sub-09 extraction was stopped. Decoders were offline, linear, and used log high-gamma only. The result supports reporting envelope-only and within-speech shape metrics alongside standard spectrogram r; above-null shape does not establish useful word decoding.

speech brain-computer interfaces · stereo-electroencephalography · spectrogram reconstruction · spectral shape · high-gamma · envelope decoding · shift null · Wilcoxon signed-rank test

Admitted 2 citing publications

References

4
  1. In three SEEG participants, standard spectrogram r matched envelope-only decoding; spectral shape exceeded shift nulls but did not identify words sci:sha256:0a9d6f0825e1212203d528303c1e9fcdc93508a63778aa21ac3c115f2d945c1f

    [Legacy relation: reproduces] The same pre-registered pipeline in one new participant (sub-05) gives the same pattern: standard r = envelope-only r, within-speech shape above the shift null, word identity from the envelope and not from the shape.

  2. usesCausal neural context yielded higher speech spectrogram scores in three SEEG participants sci:sha256:f87a783ff90af05ec65a5dd0e510c1e6bac04df76336cb3c05a5ab9d1ff6af33

    The high-gamma and log-mel feature functions (features_ref.py) are copied without change from 7748198c.

  3. supportsIn a fourth participant, SEEG spectrogram r matched envelope-only r, while within-speech shape exceeded a shift null sci:sha256:cfddfc65f6b3ac410562e59bfbf6b8cd03939a10bc0f3956b28ad044c03eff61

    In the new participant sub-05, M0 = M1 and M3 > null again. The group Wilcoxon test that b141a85d could not do (n = 4) gives p = 0.031 at n = 5.

  4. usesIn a fourth participant, SEEG spectrogram r matched envelope-only r, while within-speech shape exceeded a shift null sci:sha256:cfddfc65f6b3ac410562e59bfbf6b8cd03939a10bc0f3956b28ad044c03eff61

    All analysis code (fetch, features, decoders, metrics, null, summary) is a byte copy of the b141a85d code. The sub-04 results are taken from b141a85d.

Cited by

2 publications

Public catalog entries on Sciscape.