ggfox00000 commited on
Commit
254bfb1
·
verified ·
1 Parent(s): dd9b921

Mirror SUMM-RE test split (124 tracks ≈ 41h, byte-exact from linagora/SUMM-RE)

Browse files
README.md ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-sa-4.0
3
+ task_categories:
4
+ - automatic-speech-recognition
5
+ - voice-activity-detection
6
+ language:
7
+ - fr
8
+ size_categories:
9
+ - n<1K
10
+ pretty_name: SUMM-RE French — test split
11
+ tags:
12
+ - summ-re
13
+ - linagora
14
+ - french
15
+ - asr
16
+ - speech
17
+ - meetings
18
+ - conversational
19
+ - long-form
20
+ - vad
21
+ annotations_creators:
22
+ - expert-generated
23
+ source_datasets:
24
+ - extended|summ-re
25
+ dataset_info:
26
+ features:
27
+ - name: meeting_id
28
+ dtype: string
29
+ - name: speaker_id
30
+ dtype: string
31
+ - name: audio_id
32
+ dtype: string
33
+ - name: audio
34
+ dtype: audio
35
+ - name: segments
36
+ list:
37
+ - name: end
38
+ dtype: float64
39
+ - name: start
40
+ dtype: float64
41
+ - name: transcript
42
+ dtype: string
43
+ - name: words
44
+ list:
45
+ - name: end
46
+ dtype: float64
47
+ - name: start
48
+ dtype: float64
49
+ - name: word
50
+ dtype: string
51
+ - name: transcript
52
+ dtype: string
53
+ splits:
54
+ - name: test
55
+ num_examples: 124
56
+ num_bytes: 13775584735
57
+ configs:
58
+ - config_name: default
59
+ data_files:
60
+ - split: test
61
+ path: data/test-*.parquet
62
+ ---
63
+
64
+ # SUMM-RE — French test split (mirror of `linagora/SUMM-RE`)
65
+
66
+ Mirror **public** du split **`test` de SUMM-RE** (LINAGORA / Aix-Marseille
67
+ LPL), pour benchmark **ASR français conversationnel** (parole de réunion,
68
+ 3-4 locuteurs, ~20 min par session).
69
+
70
+ > ⚠ Ce repo ne contient **que le split test** (124 tracks individuelles =
71
+ > 37 réunions × 3-4 micros). Pour les splits `train` / `dev`, voir le repo
72
+ > upstream `linagora/SUMM-RE`.
73
+
74
+ ## Contenu
75
+
76
+ - **124** pistes audio individuelles (1 piste = 1 microphone d'un locuteur
77
+ pour une réunion entière de ~20 min)
78
+ - **37** réunions distinctes (chacune avec 3-4 micros)
79
+ - Audio : **WAV 16 kHz mono PCM_16** embarqué dans parquet (tel que upstream)
80
+ - Annotations : transcription manuelle + alignement forcé au mot
81
+ - Langue : **français (fr)**, parole conversationnelle / réunion
82
+ - Licence : **CC-BY-SA-4.0** (héritée de SUMM-RE upstream)
83
+ - Durée totale : **~41 h** (124 × ~20 min)
84
+
85
+ ## Colonnes
86
+
87
+ | Colonne | Type | Description |
88
+ |---------|------|-------------|
89
+ | `audio` | Audio | dict `{{path, array, sampling_rate}}` — décodé auto |
90
+ | `meeting_id` | string | ID réunion : `<num><letter>_<scenario><type><loc>` (ex `001a_PARL`) |
91
+ | `speaker_id` | string | identifiant locuteur |
92
+ | `audio_id` | string | `meeting_id + speaker_id` (clé unique) |
93
+ | `transcript` | string | transcription concaténée de tous les segments (ref WER) |
94
+ | `segments` | list[struct] | segments avec `start`, `end`, `transcript`, `words[start,end,word]` |
95
+
96
+ ## Utilisation
97
+
98
+ ### Chargement
99
+ ```python
100
+ from datasets import load_dataset
101
+
102
+ ds = load_dataset("ggfox00000/stt-summre-fr-test", split="test")
103
+ print(ds)
104
+ sample = ds[0]
105
+ print(sample["meeting_id"], sample["speaker_id"])
106
+ print(sample["audio"]["sampling_rate"], sample["audio"]["array"].shape)
107
+ print("segments:", len(sample["segments"]))
108
+ print("first seg:", sample["segments"][0])
109
+ ```
110
+
111
+ ### Découpage par segment pour évaluation WER
112
+ SUMM-RE livre **1 ligne par piste audio complète (~20 min)**. Pour comparer
113
+ à un benchmark utterance-level, tu peux soit :
114
+
115
+ 1. **Long-form WER** : passer la track complète à Whisper en mode
116
+ chunked (`chunk_length_s=30`, `return_timestamps=True`), comparer le
117
+ résultat au champ `transcript`.
118
+ 2. **Utterance-level** : découper par `segments[i].start/end`, transcrire
119
+ chaque segment isolément, comparer au `segments[i].transcript`.
120
+
121
+ ### Évaluation diarisation
122
+ En fusionnant toutes les tracks d'une même `meeting_id`, tu peux évaluer
123
+ de la diarisation (qui parle quand). Les timestamps `segments[i].start/end`
124
+ servent de RTTM de référence. Voir paper Hunter et al. (2024) §5.
125
+
126
+ ## Pré-traitement
127
+
128
+ **Aucun.** Les 28 parquets sont transférés bit-à-bit depuis `linagora/SUMM-RE`
129
+ (uniquement les fichiers du split `test`, l'arbo `data/test/test-*.parquet`
130
+ upstream est aplatie en `data/test-*.parquet` pour cohérence avec nos autres
131
+ STT datasets `ggfox00000/stt-*`). Aucun resampling, aucun ré-encodage.
132
+
133
+ ## Source
134
+
135
+ - Dataset upstream : https://huggingface.co/datasets/linagora/SUMM-RE
136
+ - Paper : Hunter et al. 2024, *"SUMM-RE: A corpus of French meeting-style
137
+ conversations"* (LREC-COLING 2024)
138
+ - Projet : SUMM-RE (ANR-20-CE23-0017), LINAGORA + Aix-Marseille LPL
139
+
140
+ ## Licence
141
+
142
+ **CC-BY-SA-4.0** (héritée de SUMM-RE upstream). Toute redistribution doit
143
+ préserver la licence et créditer LINAGORA + le LPL.
144
+
145
+ ## Citation
146
+
147
+ ```bibtex
148
+ @inproceedings{{hunter2024summre,
149
+ title = {{SUMM-RE: A Corpus of French Meeting-Style Conversations}},
150
+ author = {{Hunter, Julie and others}},
151
+ booktitle = {{Proceedings of LREC-COLING 2024}},
152
+ year = {{2024}},
153
+ }}
154
+ ```
155
+
156
+ Pour le pipeline d'alignement Whisper utilisé sur le split `train` (non
157
+ inclus dans ce mirror) :
158
+
159
+ ```bibtex
160
+ @inproceedings{{yamasaki2023transcribing,
161
+ title = {{Transcribing And Aligning Conversational Speech: A Hybrid
162
+ Pipeline Applied To French Conversations}},
163
+ author = {{Yamasaki, Hugo and others}},
164
+ booktitle = {{Proceedings of ASRU 2023}},
165
+ year = {{2023}},
166
+ }}
167
+ ```
data/test-00000-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66a24b8684965648bd35dc00ef754ae35c3ff1c5d2c02bdfb4c5e0a0c597f735
3
+ size 563896371
data/test-00001-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07f61b0dda52940e407b7594961fc458942ecb668d4b542f362a967d1719b5c4
3
+ size 523086010
data/test-00002-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba6ceb4c9556d2b17425846f55f367af050d979ed80d73054d99bdb1638a01a6
3
+ size 459001684
data/test-00003-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e747ea1e72878d1dbc7b02b2e34a0422bf64ae0066ffb437cb4a0cc0d7165846
3
+ size 482580189
data/test-00004-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8840ac610342276cea8982268a1b1a5d258a4b15b5f5402970dd5d732e8d71b
3
+ size 447942878
data/test-00005-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c3e0b877a0e92f4443fbd5d2ba0e65f72eb67fd0410a97178bbcf910aab809c9
3
+ size 557470205
data/test-00006-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25ac428e3cd87da2b499293a0391aa6dd4867379e930a9e323f27ff85fe215da
3
+ size 579387239
data/test-00007-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:32a88df973e2e0f08b293b5f7b7150ad6a2f52371e8d1400b69836590f6ace86
3
+ size 621881054
data/test-00008-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3c71d9ddb154ce3187fbda46e3b0b47e52484eff4b85ce068593c6e0c1fc58d
3
+ size 531195710
data/test-00009-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68ff36ea9f3c3f36b17ca81b1e274e8b15259bd80e0927b14cc3884b47ab0d21
3
+ size 547498778
data/test-00010-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e5647b27537dc68d1a7594ea46ad6404c1953a84adeb3365eb92c820e5ee29cf
3
+ size 526807885
data/test-00011-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ad86283302a50fa6a7ec499419f2cd1a34d8a3ba8bbf00aff71106630220b88
3
+ size 589598900
data/test-00012-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2ca56639acd7c4001a7863b701a066dfec76fb1e4e91b0acadd8b4d3c2892b4
3
+ size 366628372
data/test-00013-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39058e97d5d66d6fb123cb7bbb2ceed80724aa6063a165824e95c9d18098a6ee
3
+ size 372575232
data/test-00014-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e519d2a77d585741d5b55139aed4cfe08e78ee326400ace669954df44d7e79e
3
+ size 429522513
data/test-00015-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f3d62b262d794ae5deaa1daba2e0b6ab4c9262030a6e6c77cd445dc25799fd4
3
+ size 464383945
data/test-00016-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6077ab0c8b78d8e70b3b8ec743f3d9f4ce4c80e59fa55167c2a59ee008b15dc9
3
+ size 518893068
data/test-00017-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85f1c4887e91b278924da70573f637f1025cc76408035aa018103960760be3d6
3
+ size 357909169
data/test-00018-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:216fb990fc052a5aa6e9633385a14716503d07915004b03ca6e673caee12af52
3
+ size 349920433
data/test-00019-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c35de629ad945a3f2eddedcc55be7d8904939231db558c9cd078280816d381a
3
+ size 405521195
data/test-00020-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8f00609af2e8fe1d28832c96e8f6702e7d87da667911d6739ebe8ebe1b8812b
3
+ size 383233919
data/test-00021-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:850e6c325ebd385ffcd948cadcb54a8285cb340c4e134181ce4fd550042338c3
3
+ size 180686722
data/test-00022-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ac3817c89e7683af45580e86a781a716d72391139c9e81e28cbabbd8c5b218c
3
+ size 423421100
data/test-00023-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dca890a79f21b7030916798c3160b42390524ab1673627b80b967299227f699b
3
+ size 478910647
data/test-00024-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40b43b0763ddcaa610559a51e4a4d89ac87d2822ba6c367335be4eab556ec1c0
3
+ size 312298534
data/test-00025-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:032715fd412cbbde0781815f008717abbf1514c591183cd2bb3b7ce202e1eb4a
3
+ size 317377955
data/test-00026-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e387d2f2fb2eab983650ae6ea4bac4113a339ffcd1158444c2ac1dff7bf193f7
3
+ size 446756284
data/test-00027-of-00028.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d30f77c06d9b19692c379fead77f73fc709d15e488c8d6e86f674be27214277b
3
+ size 485900035