Datasets:

Tasks:
Other
Modalities:
Audio
Size:
< 1K
ArXiv:
Libraries:
Datasets
License:
oiwi commited on
Commit
4ffb27d
·
verified ·
1 Parent(s): 017e3c8

Add Sherlock 1 stimulus README

Browse files
Files changed (1) hide show
  1. Sherlock1/stimuli/README.md +78 -0
Sherlock1/stimuli/README.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # A Study in Scarlet audio stimuli
2
+
3
+ This directory contains the LibriVox audiobook audio used as auditory stimuli
4
+ for Sherlock 1 in LibriBrain100.
5
+
6
+ ## Source
7
+
8
+ The stimulus material is *A Study in Scarlet* by Arthur Conan Doyle, read by
9
+ David Clarke for LibriVox.
10
+
11
+ LibriVox source:
12
+
13
+ https://librivox.org/a-study-in-scarlet-version-6-by-sir-arthur-conan-doyle/
14
+
15
+ The corresponding text was obtained from Project Gutenberg.
16
+
17
+ Both the text and the LibriVox recording used here are public-domain material.
18
+
19
+ ## Files
20
+
21
+ The `audio/` directory contains the two WAV files used for the Sherlock 1
22
+ validation and test sessions in LibriBrain100:
23
+
24
+ - `studyinscarlet_11_doyle_64kb.wav` - Chapter 11, *A Flight For Life*
25
+ - `studyinscarlet_12_doyle_64kb.wav` - Chapter 12, *The Avenging Angels*
26
+
27
+ These stimuli were presented to participants `sub-1` through `sub-32`.
28
+
29
+ The same stimulus WAV files were previously released in the original
30
+ LibriBrain dataset for `sub-0`. They are reproduced here so that LibriBrain100
31
+ contains the complete set of stimuli corresponding to its MEG recordings.
32
+
33
+ ## Stimulus preparation
34
+
35
+ For the LibriBrain experiment, the source audiobook audio was converted to
36
+ uncompressed WAV format and resampled to 48 kHz using SoX.
37
+
38
+ The audio was segmented into natural sentences or phrases using voice
39
+ activity detection followed by manual review and correction.
40
+
41
+ The corresponding text was obtained from Project Gutenberg and then manually
42
+ corrected to match the spoken audiobook. Text was also normalised to reflect
43
+ the spoken form, for example by spelling out numbers and resolving other
44
+ written forms that were ambiguous or differed from the narration.
45
+
46
+ ## Linguistic annotations
47
+
48
+ Time-aligned linguistic annotations are provided in the `*_events.tsv` files
49
+ alongside the corresponding MEG recordings in the dataset.
50
+
51
+ These include word- and phoneme-level annotations aligned to the stimulus
52
+ audio.
53
+
54
+ ## Dataset split
55
+
56
+ The two Sherlock 1 sessions form the held-out validation and test material
57
+ used by the standard LibriBrain100 data loaders and benchmark:
58
+
59
+ - Chapter 11, *A Flight For Life*: validation
60
+ - Chapter 12, *The Avenging Angels*: test
61
+
62
+ ## Licence
63
+
64
+ The audiobook recording and source text used here are public-domain
65
+ materials.
66
+
67
+ The LibriBrain100-specific annotations and dataset organisation are released
68
+ under the applicable LibriBrain100 dataset terms.
69
+
70
+ ## Citation
71
+
72
+ If you use these stimuli together with the corresponding MEG data, please
73
+ cite the LibriBrain100 dataset paper.
74
+
75
+ Please also credit the original LibriVox recording and reader:
76
+
77
+ Arthur Conan Doyle, *A Study in Scarlet*.
78
+ Read by David Clarke. LibriVox.