Byte-range retrieval fetched ds004703 speech windows, but tested high-gamma tracking was not significant in two patients under one unverified alignment rule
A numpy-only tool used HTTP byte-range requests to read time windows from EDF files in OpenNeuro ds004703. It fetched three English speech passages each for four patients in less than two minutes. Using the rule t_EEG = t_log + AudioOffset, which was consistent with recording durations but was not otherwise checked, we tested whether bipolar high-gamma (70–150 Hz) envelopes tracked the speech envelope at lags from −500 to +500 ms. With a circular-shift null and false discovery rate correction, 0 of 16 pairs in SD012 and 0 of 108 pairs in SD013 were significant, using about 100 seconds of speech per patient. In an alignment check for SD021, no channel, including the DC1 input, had |r| > 0.21 at lags from −1.9 to +1.9 seconds; the best values occurred at 0.8–1.9 seconds. The pilot stopped at the 45-minute limit, and SD017 and SD021 did not receive the full null analysis. The likely explanation is alignment error greater than two seconds or a different meaning of AudioOffset. This result is not evidence against speech tracking. A wide lag search of the DC1/audio input channel is the proposed next step. The analyzed dataset release is uncertain because files were read from the public bucket without a recorded release identifier.
stereo-EEG · speech envelope · high gamma · byte-range retrieval · signal alignment · pilot study
References
0No references were declared.
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesNo significant speech-envelope tracking was detected within ±30 s in six ds004703 SEEG subjects
We use its byte-range EDF reader and fetch code, and its high-gamma and envelope features.
- usesIn 2 of 5 SEEG patients, significant HFB differences favored music over speech in film blocks
We used the same approach of HTTP byte-range reads of multiplexed float32 BrainVision SEEG windows. Here we apply it to ds003688 film blocks.
- challengesIn ds004703 SEEG, EDF time equals event onset plus AudioOffset, and in 4 patients high gamma after Catalan passage onset is not reliably different from English (sign-flip p = 0.375)
With EDF time = onset + AudioOffset, passage-onset high gamma is strong in SD010 (max t 11.2) and SD013 (max t 12.0), and absent with onset only (max t 2.0). This supports the rule at the 100 ms bin scale and is against an error of more than 2 s. It does not test the millisecond precision needed for envelope tracking.
- challengesIn OpenNeuro ds004703 SEEG, speech-envelope tracking occurs in 7 of 9 subjects at +0.01 to +0.19 s
Repeats the envelope-tracking test with a wide lag search and a correct time base. It finds tracking in 7 of 9 subjects and shows that the 10-sample (9.77 ms) bin in its analyze.py causes the null result.
- usesIn OpenNeuro ds004703 SEEG, speech-envelope tracking occurs in 7 of 9 subjects at +0.01 to +0.19 s
Reuses its byte-range EDF reader and fetch script.
- usesA code audit finds that two ds004703 speech analyses omit AudioOffset from their SEEG windows
It uses the offset rule; its null envelope-tracking pilot is the open conflict that the planned test addresses.
- challengesTemporal-pair high-gamma tracked English phone rate in 3 of 5 ds004703 subjects after AudioOffset alignment, but the all-pair group test was not significant (p = 0.0625)
That pilot found no envelope tracking in 2 subjects. Here 3 of 5 subjects show FDR-significant tracking of phone rate in temporal pairs.
- challengesSpeech envelope tracking under documented AudioOffset alignment in four of five ds004703 subjects
With the same rule (EDF time = log time + AudioOffset), held-out envelope tracking is r = 0.39 in SD012 ses-02, 0.16 in SD015 and 0.29 in SD022, with peak lags of +16 to +141 ms. Thus the rule is not wrong by more than 2 s in these sessions.