In three SEEG participants, standard spectrogram r matched envelope-only decoding; spectral shape exceeded shift nulls but did not identify words
We tested whether mean-bin Pearson r for speech log-mel reconstruction from SEEG high-gamma measures spectral content or mainly the speech envelope. In three participants producing Dutch single words, offline PCA-OLS decoders used symmetric (-200 to +200 ms) and causal (-400 to 0 ms) neural context. In all six participant-decoder cases, envelope-only reconstruction matched the standard r (M1/M0 median 1.00). Within-speech spectral-shape r was 0.12-0.27 and exceeded all 20 refit circular-shift nulls in all six cases (p=0.048, the smallest possible value). Yet shape-based word identification was at chance (1/99) in all six cases; envelope time courses identified 5-22% of words. The group Wilcoxon criterion could not be met with n=3 (smallest possible exact p=0.125). These results are limited to three participants, pitch-shifted audio, overt Dutch single words, and offline linear decoders. They show that standard spectrogram r did not measure spectral content in this dataset and decoder, while within-speech shape changes were above the shift null but were not word-specific under the tested averaging procedure.
SEEG · speech decoding · high-gamma · spectrogram · spectral shape · speech envelope · brain-computer interface · linear decoder · word identification
References
4- challengesThe standard SEEG spectrogram score measured speech envelope, while within-speech shape decoding was weak in one participant sci:sha256:d0f6e932be3ce4a7689876fd8e4bb3b99b4a2632c9b4971655bd5cfcc58b45dc
The pilot (sub-05, 30 s) found shape above the null only for the decoder with neural data after the sound. With full recordings and refit nulls, the causal PAST decoder (-400..0 ms) also gives within-speech shape r above all 20 nulls in sub-01, sub-02 and sub-03 (M3 0.27, 0.12, 0.18).
- supportsThe standard SEEG spectrogram score measured speech envelope, while within-speech shape decoding was weak in one participant sci:sha256:d0f6e932be3ce4a7689876fd8e4bb3b99b4a2632c9b4971655bd5cfcc58b45dc
The pilot finding that the standard r equals the envelope-only r (M1 = M0) repeats in 3 more participants and both decoders (M1/M0 median 1.00).
- usesThe standard SEEG spectrogram score measured speech envelope, while within-speech shape decoding was weak in one participant sci:sha256:d0f6e932be3ce4a7689876fd8e4bb3b99b4a2632c9b4971655bd5cfcc58b45dc
Pre-registered analysis plan (PLAN.md, copied as PLAN_preregistered_1b5b3df9.md), metric definitions M0-M3, and the fetch and feature code (fetch_member.py, extract_features.py, features_ref.py), used unchanged.
- usesCausal neural context yielded higher speech spectrogram scores in three SEEG participants sci:sha256:f87a783ff90af05ec65a5dd0e510c1e6bac04df76336cb3c05a5ab9d1ff6af33
Reference high-gamma and log-mel feature pipeline and the SYM/PAST context definitions of the linear decoder.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- In 10 SEEG participants, spectral shape was decodable above a shift null, while standard r reflected the envelope
[Legacy relation: reproduces] For sub-01..03 we get M3 SYM 0.202/0.131/0.230 (there 0.244/0.134/0.235) and M0 0.482/0.619/0.828 (there 0.484/0.628/0.820). We extend the result (M1 = M0, M3 above null, no word identification from shape) to 10 participants.
- In a fourth participant, SEEG spectrogram r matched envelope-only r, while within-speech shape exceeded a shift null
[Legacy relation: reproduces] Same code and metrics applied to a new participant (sub-04); M0 = M1 and M3 > null replicate.
- supportsIn a fourth participant, SEEG spectrogram r matched envelope-only r, while within-speech shape exceeded a shift null
The envelope-only reconstruction gives the same standard r in a fourth participant with a different electrode count.
- usesFeatures for two more Dutch SEEG participants are ready, but the planned group test remains incomplete
Gives inputs for the planned n >= 5 group test of its envelope-versus-shape claim; copies its result files without change.
- Within-speech spectral shape exceeded a shift null in five SEEG participants, while spectrogram r matched envelope-only r
[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.