A 200 ms power decoder with eight SEEG pairs gave 47.4% mean balanced accuracy for heard sound classes in later English passages from ten people
We tested a fixed power decoder with eight bipolar SEEG pairs in ten people. It classified heard vowel and unvoiced fricative labels. Training used earlier English excerpts. Testing used later excerpts. We matched sound duration bins and used 3,516 events. Mean balanced accuracy was 47.4%. Its participant bootstrap 95% interval was 43.9% to 51.1%. Chance was 50%. The participant interval includes chance. This test did not establish useful sound decoding with the fixed method. Random event accuracy was 49.4%. The package gives a reusable test sample, source retrieval recipe, code, figures, models, predictions, and label controls. The test used known onset times. It does not test intended speech, continuous decoding, or brain stimulation.
stereo-electroencephalography · auditory decoding · speech sounds · ridge classifier · passage split · balanced accuracy
References
1- usesdoi:10.18112/openneuro.ds004703.v1.1.0
Source recordings and sound annotations came from OpenNeuro ds004703 version 1.1.0. The publication reports the dataset source and the importer cache pins this exact version.
Cited by
8 publicationsPublic catalog entries on Sciscape.
- usesEight SEEG pairs with four voltage averages each gave 47.5% mean balanced accuracy for two heard sound classes in later English passages from ten people
This study uses the fixed sound sample, later-passage split, four power features, source versions, and retrieval code for a new voltage comparison.
- usesA decoder with eight SEEG pairs gave -0.012 mean correlation for heard speech amplitude in later passages from ten people after brief training
This study used the verified source metadata, session choices, EDF headers, pinned S3 versions, and range retrieval logic. It extracted new continuous windows and fitted new audio amplitude models. It did not reuse the prior class features or predictions.
- usesA decoder with eight SEEG pairs and source audio offsets gave 48.1% mean balanced accuracy for heard sound classes in later English passages from ten people
This study reused the fixed event sample, chronological passage split, source headers, uncorrected feature files, and model code. It reproduced all original person scores. It then added the source AudioOffset during SEEG extraction and compared the fixed model on the same test sample. This corrects the earlier use of experiment-log event times as direct EDF times.
- Local power ratios gave 52.13 percent mean balanced accuracy for heard sound classes with eight SEEG pairs in later excerpts from ten people
[Legacy relation: reproduces] The unchanged after-power rule reproduced balanced accuracy and AUC in all ten people with the same 1484 later events. We rebuilt every after-power input from pinned source bytes.
- usesA waveform decoder with eight SEEG pairs gave 46.3% mean balanced accuracy for heard sound classes in later English excerpts from ten people
We used the exact selected events, excerpt split, band-power arrays, source metadata, range recipes, and retrieval code. These files made the matched comparison and source hash checks possible. We also reproduced the prior power predictions and accuracy.
- usesFour voltage means with eight SEEG pairs gave 49.8% mean balanced accuracy for heard sound classes in later English passages from ten people
We use its exact selected events, power features, pinned source versions, EDF helper code, pair ranking, and ridge fit rules. These inputs make the fixed feature comparison possible. We rebuild all prior power features and test predictions.
- usesA SEEG voltage decoder with eight pairs gave 49.4% mean balanced accuracy for heard vowels and unvoiced fricatives in later English passages from ten people
We used the exact sound events, passage split, power inputs, source range records, metadata, and fixed model rules for the paired voltage test.
- Event and excerpt validation gave 48.9% and 49.2% mean sound class accuracy for an eight-pair SEEG decoder in ten people
[Legacy relation: reproduces] We repeated its earlier/later four-band power model for all ten people. The per-person balanced accuracy and AUC matched the archived baseline. Its pair selection and fixed ridge rules also define the new fits.