Emulated-Inc/meg-story-phoneme-decoding

MEG-MASC: preprocessed MEG with per-sample phonetic labels

This repository is a mirror for reproducible builds: it holds the files produced once from the published originals listed below by the processing described here, so that later builds get exactly the same bytes (SHA256SUMS gives the SHA-256 of each file). Please cite and credit the original authors.

Contents

Uncompressed tar files (agent-*.tar) that unpack to:

• train/<subject>_ses-<k>_<story>/: 22 subjects, three stories (cable_spool_fort, easy_money, the_black_willow): meg.npy (float16 [224, samples]: the 208 MEG channels and 16 reference magnetometers in femtotesla, high-passed at 0.1 Hz, decimated from 1000 Hz to 100 Hz, clipped to +-60000), labels.npy (int8 [samples, 15]: 14 phonetic features and word onset; -1 where undefined), events.tsv (phoneme and word events) and info.json;
• audio/<story>_<k>.wav: the stories' sound files;
• dev/<subject>_ses-<k>/: segments easy_money_6 and easy_money_7 cut from the easy_money recordings.

The recordings of the story lw1 and of subjects sub-02, sub-08, sub-13, sub-19 and sub-24 are not included.

Total: 7.34 GB. Check a download with sha256sum -c SHA256SUMS.

file	bytes	sha256
agent-000.tar	2,000,056,320	ea7169dc8ec491e28d8896d6db37b64500b09e5424f1fc85cf467669176257d5
agent-001.tar	1,987,246,080	10dfd89991f8399ce04339ae0569790e233cfff9b7266f2a834109db33e02607
agent-002.tar	1,980,211,200	9c6acd4a0eb1d960b9bbb535cc89065e9e9aa580d7fb5f52340860e72f376ce9
agent-003.tar	1,369,784,320	e6ba4db6a30ac1365125cb764417675b8f284a1807cd1dfc3051f2cb9c3143b8

Source

• L. Gwilliams, G. Flick, A. Marantz, L. Pylkkänen, D. Poeppel and J.-R. King, "Introducing MEG-MASC a high-quality magneto-encephalography dataset for evaluating natural speech processing", Scientific Data 10, 862 (2023). Data: https://osf.io/ag3kj/ (with its components h2tzn, u5327 and dr4wy). The stories are from the Manually Annotated Sub-Corpus (MASC) of the Open American National Corpus, as distributed with MEG-MASC.

Licence

The MEG recordings of MEG-MASC are released under the Creative Commons CC0 1.0 Universal public-domain dedication; the processed MEG arrays here are shared under CC0 1.0 as well. The story texts are from MASC (American National Corpus project), licensed CC BY 3.0 US; MEG-MASC distributes synthesised readings of them. The story audio (audio/*.wav) and the word and phoneme annotations derived from the texts (events.tsv, labels.npy) are shared under CC BY 3.0 US.

Attribution: MEG-MASC by L. Gwilliams, G. Flick, A. Marantz, L. Pylkkänen, D. Poeppel and J.-R. King (CC0 1.0); story texts from MASC, American National Corpus project (CC BY 3.0 US). Filtered, decimated, cropped and labelled.

No endorsement by the original authors is implied.