Hugging Face's logo

Datasets:
SMIIP-lab
/
AISHELL8-RealScene

Modalities:
Audio
Tags:
audio
speech
chinese
mandarin
audiovisual
avsr
License:
Dataset card Data Studio Files
xet
 Community
SMIIP-lab commited on
Commit
bdab0bf
·
verified ·
1 Parent(s): 4514234

Add files using upload-large-folder tool

Browse files
Files changed (20) hide show
  1. .gitattributes +0 -1
  2. README.md +208 -3
  3. SHA256SUMS +34 -0
  4. audio/dev_far_audio.tar.gz +3 -0
  5. audio/dev_near_audio.tar.gz +3 -0
  6. audio/eval_far_audio.tar.gz +3 -0
  7. audio/eval_near_audio.tar.gz +3 -0
  8. audio/train_far_audio_00.tar.gz +3 -0
  9. audio/train_far_audio_01.tar.gz +3 -0
  10. audio/train_far_audio_02.tar.gz +3 -0
  11. audio/train_far_audio_03.tar.gz +3 -0
  12. audio/train_far_audio_04.tar.gz +3 -0
  13. audio/train_far_audio_05.tar.gz +3 -0
  14. audio/train_near_audio_00.tar.gz +3 -0
  15. audio/train_near_audio_01.tar.gz +3 -0
  16. audio/train_near_audio_02.tar.gz +3 -0
  17. audio/train_near_audio_03.tar.gz +3 -0
  18. audio/train_near_audio_04.tar.gz +3 -0
  19. upload_aishell8_realscene_resume.log +0 -0
  20. video/train_video_10.tar.gz +3 -0
.gitattributes CHANGED
@@ -58,4 +58,3 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
- upload_aishell8_realscene.log filter=lfs diff=lfs merge=lfs -text
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
README.md CHANGED
@@ -1,3 +1,208 @@
1
- ---
2
- license: cc-by-nc-sa-4.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - audio
4
+ - speech
5
+ - chinese
6
+ - mandarin
7
+ - audiovisual
8
+ - avsr
9
+ - multimodal
10
+ - far-field
11
+ - multi-view
12
+ - conversation
13
+
14
+ license: cc-by-nc-sa-4.0
15
+
16
+ extra_gated_heading: "Data Application"
17
+
18
+ extra_gated_description: |
19
+ Please fill out the form below to apply for access to this dataset.
20
+ Access will be granted only for academic and non-commercial research use.
21
+ Please use your institutional email address and agree to the dataset license terms.
22
+
23
+ extra_gated_button_content: "Submit Application"
24
+
25
+ extra_gated_fields:
26
+ First Name:
27
+ type: text
28
+ required: true
29
+ Last Name:
30
+ type: text
31
+ required: true
32
+ Affiliation (Research organization):
33
+ type: text
34
+ required: true
35
+ Email Address (Please use institutional e-mail):
36
+ type: text
37
+ required: true
38
+ I have read and agree to the license terms (CC BY-NC-SA 4.0):
39
+ type: checkbox
40
+ required: true
41
+ ---
42
+
43
+ ## 📘 Dataset Summary
44
+
45
+ | **Property** | **Description** |
46
+ | :------------------- | :-------------------------- |
47
+ | **Language** | Chinese (Mandarin, ZH) |
48
+ | **License** | CC BY-NC-SA 4.0 |
49
+ | **Total Duration** | 102.19 hours |
50
+ | **Speakers** | 171 foreground speakers |
51
+ | **Scenes** | 5 real-world locations |
52
+ | **Recording Style** | Conversational speech |
53
+ | **Audio** | Near-field + far-field |
54
+ | **Video** | Multi-view RGB facial video |
55
+ | **Sampling Rate** | 16 kHz |
56
+ | **Far-field Audio** | 8-channel |
57
+ | **Video Resolution** | 256×256 @ 25 fps |
58
+
59
+ ---
60
+
61
+ ## 🎙️ Dataset Description
62
+
63
+ AISHELL8-RealScene is a **public multi-scenario and multi-view audio-visual conversational corpus** recorded in real-world environments.
64
+
65
+ Unlike most existing AVSR datasets collected under constrained frontal settings, AISHELL8-RealScene focuses on realistic conditions including:
66
+
67
+ * environmental noise
68
+ * viewpoint variation
69
+ * speaker interference
70
+ * background activities
71
+ * outdoor scenes
72
+ * conversational interactions
73
+
74
+ The corpus was collected by AISHELL using a synchronized multi-device recording setup.
75
+
76
+ ---
77
+
78
+ ## 🏛️ Recording Environment
79
+
80
+ AISHELL8-RealScene contains recordings from **five real-world locations**:
81
+
82
+ | ID | Location | Type |
83
+ | -- | -------- | ------- |
84
+ | L1 | Building | Outdoor |
85
+ | L2 | Hall | Indoor |
86
+ | L3 | Hotel | Indoor |
87
+ | L4 | Park | Outdoor |
88
+ | L5 | Street | Outdoor |
89
+
90
+ The recordings cover both indoor and outdoor scenarios.
91
+
92
+ ---
93
+
94
+ ## 🔊 Audio Specifications
95
+
96
+ AISHELL8-RealScene provides both:
97
+
98
+ * **near-field audio**
99
+ * **far-field microphone-array audio**
100
+
101
+ ### Near-field Audio
102
+
103
+ Each foreground speaker wears a close-talking microphone used for transcription and alignment.
104
+
105
+ ### Far-field Audio
106
+
107
+ Far-field audio is recorded using:
108
+
109
+ * circular microphone array
110
+ * 16 microphones
111
+ * 16 kHz
112
+ * 16-bit
113
+
114
+ To reduce release size while preserving spatial information, the public release provides:
115
+
116
+ * **8-channel far-field audio**
117
+
118
+ obtained by uniformly selecting microphones from the array.
119
+
120
+ ### Audio Format
121
+
122
+ * Format: `.wav`
123
+ * Sampling rate: **16 kHz**
124
+ * Near-field: single-channel
125
+ * Far-field: 8-channel
126
+
127
+ ---
128
+
129
+ ## 🎥 Video Specifications
130
+
131
+ Video is captured using **three synchronized HD RGB cameras**.
132
+
133
+ ### Camera Views
134
+
135
+ The three cameras are positioned horizontally with approximately **30° angular intervals**:
136
+
137
+ * **D0** — left
138
+ * **D1** — center
139
+ * **D2** — right
140
+
141
+ ### Video Parameters
142
+
143
+ | Property | Value |
144
+ | ---------- | ------------ |
145
+ | Resolution | 256×256 |
146
+ | Frame Rate | 25 fps |
147
+ | Views | D0 / D1 / D2 |
148
+
149
+ The released facial videos are:
150
+
151
+ * resized to **256×256**
152
+ * synchronized with audio
153
+
154
+ ---
155
+
156
+ ## 📊 Dataset Statistics
157
+
158
+ ### Dataset Split
159
+
160
+ | Split | Duration (h) | Sessions | Groups | Speakers |
161
+ | ----- | -----------: | -------: | -----: | -------: |
162
+ | Train | 79.84 | 133 | 56 | 133 |
163
+ | Dev | 10.70 | 18 | 7 | 18 |
164
+ | Eval | 11.65 | 20 | 7 | 20 |
165
+ | Total | 102.19 | 171 | 70 | 171 |
166
+
167
+ No speakers overlap across:
168
+
169
+ * train
170
+ * dev
171
+ * eval
172
+
173
+ All splits contain recordings from all five locations.
174
+
175
+ ---
176
+
177
+
178
+
179
+ ## 📜 License
180
+
181
+ This dataset is released under:
182
+
183
+ **CC BY-NC-SA 4.0**
184
+
185
+ https://creativecommons.org/licenses/by-nc-sa/4.0/
186
+
187
+ ---
188
+
189
+ ## 📄 Dataset Resources
190
+
191
+ - 📄 **Paper:** [arXiv:](https://arxiv.org/abs/)
192
+ - 🌐 **Demo:** [xxx.github.io/AISHELL8-RealScene](https://xxx.github.io/AISHELL8-RealScene/)
193
+
194
+
195
+ ---
196
+
197
+ ## 📜 Citation
198
+
199
+ If you use AISHELL8-RealScene, please cite:
200
+
201
+ ```bibtex
202
+ @article{su2026m2savsr,
203
+ title = {M2S-AVSR: Modality-aware Multi-view Self-supervised Representation for Robust Audio-Visual Speech Recognition},
204
+ author = {Su, Fei and Li, Cancan and Li, Ming and Liu, Juan},
205
+ journal = {},
206
+ year = {2026}
207
+ }
208
+ ```
SHA256SUMS ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ f2948228bf69a475632b2036f51ea3ea14eb069110d60989d82a4bef0bba461f audio/dev_far_audio.tar.gz
2
+ 038b50fea099a01da59ba2c1ab04390a2592a4f971944c453257d4dfe2403918 audio/dev_near_audio.tar.gz
3
+ 15319f3473a6285c0e58c97ec07c2204ae0771d618e2a237d0c4b14682d7734c audio/eval_far_audio.tar.gz
4
+ cef343c1f0049746b926e0d4a183b7b77375965f6a44e2704a349ce50f382aa1 audio/eval_near_audio.tar.gz
5
+ 1fb6df5f31a206d903db1f125c68015d69c2ea27aefb25280b093cec44770ac6 audio/train_far_audio_00.tar.gz
6
+ 160e3be4abfa0deef5668b601d5daae05fa65456f796bbb95a8bcc91f045e29e audio/train_far_audio_01.tar.gz
7
+ 313cd0aa1c0faab5f04387a24821f4d253c9cc07d89b6bab6e6e5a13900f5976 audio/train_far_audio_02.tar.gz
8
+ 60265159c7b50dc521373a4f0a7206a325c69cdb780bd8be6ef7270fddb6d210 audio/train_far_audio_03.tar.gz
9
+ be85c1c4195595dd4a111db68bf9a0a92754228647b2d777e186decec4f39eed audio/train_far_audio_04.tar.gz
10
+ d597fd6e0d7438904cddbadb26ceebdaafce6c9f6b042a6be68a45cd0a64f8ca audio/train_far_audio_05.tar.gz
11
+ 728a3ac6d9f3d90ad7303afa5d2bda9818a88d2287ab4f004054390de9bcf3dd audio/train_near_audio_00.tar.gz
12
+ c27dbcac3c96889e438c5fe30c3e2a3b63566c6e053c741e05e11c6f04825473 audio/train_near_audio_01.tar.gz
13
+ 9d602e26e4c77fab963dbd0a1d7838ffdbedf3dec8b860724a0571e3923643c3 audio/train_near_audio_02.tar.gz
14
+ 6a70ca8e5856d2dc9bcfb3a9c3cf9d680a91a45e52f246d6e8ebd3d66c5a1cea audio/train_near_audio_03.tar.gz
15
+ 70f6166466253b247ade81846e7f072498e690462a69932f89ddfe7a268a01e7 audio/train_near_audio_04.tar.gz
16
+ 6fc0178fe4231fbc63e765b3a5caf3d7532e3fbdfb8076824b25ca64ab45c24d video/dev_video_00.tar.gz
17
+ f2799b9878c1e5a2fd5c0513725644a93641883e1e3df98092708a353c46f22a video/dev_video_01.tar.gz
18
+ 9102406e31c8e8160fb59297aac37cfeef37b389cade39592d46e0a72e4bde64 video/eval_video_00.tar.gz
19
+ a8fed62d3c7957ad830405b415ff141fad6043699ee812f901a98ac1cc27387a video/eval_video_01.tar.gz
20
+ 2539adcd693c61d8217d312a9b5d17e5a4c82ce75f1c0d0479263a06da5d47b8 video/eval_video_02.tar.gz
21
+ 6796f4252e431888bab5bbd3d9b2a1c4728b5f5c1b7f06bd34391273767f9c03 video/train_video_00.tar.gz
22
+ 1db780b669767cdfe6a041c250e0b7da8f9ea461dc9e69ce39793e619b51d257 video/train_video_01.tar.gz
23
+ 547e6170fc8f20b6bf767cad9d0be6552ebb89acab9b128782d0e198dac2c88c video/train_video_02.tar.gz
24
+ 1aba910f4ef132d595438f143e0e5d51962fbc421f06418ff3738debd39210d6 video/train_video_03.tar.gz
25
+ 01bd295e263a2ea99c86f35a98212cc6d801f2ad9a5e9249a5f6c7d0f0c9667c video/train_video_04.tar.gz
26
+ 7283807554f099dded4bdb92467b3ebf8547c6e4220ccd40587f80c861a11a8b video/train_video_05.tar.gz
27
+ 2ca063a8d5090bc44cabe4c726386d4921591db9417ee1e098192b9948bdf230 video/train_video_06.tar.gz
28
+ 39c326f1e80e6bcd0241c2caf1e866a3d983f70ccfd5e17962ba2b7879e69afc video/train_video_07.tar.gz
29
+ bcec8be0933b93fc08eb7b0a2ce91ec4c6423eb49b4c697eeb6fbf92b904b98d video/train_video_08.tar.gz
30
+ fcbdee917a89fdc57b249ea10ca8af8a7c41836977887641335c2820653b2dc5 video/train_video_09.tar.gz
31
+ 7b9b292e51db06665e44209bf68bc44cd0114f3fb246f33ad4fdd27450778ce5 video/train_video_10.tar.gz
32
+ 7bcf46d4904b031e34863c60b8e210187b8fe6d0229d7708336e28af87f88635 video/train_video_11.tar.gz
33
+ eb7bdfa3bbe4d341ebb33fe0cc6cdedb4bfbd73cb8907df3ba22fbc3b327d072 video/train_video_12.tar.gz
34
+ dd23f990d8bf02ce75c1f1c9b062e6dc90749feecbd99038f186ddff80253b59 transcription.tar.gz
audio/dev_far_audio.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2948228bf69a475632b2036f51ea3ea14eb069110d60989d82a4bef0bba461f
3
+ size 8484647928
audio/dev_near_audio.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:038b50fea099a01da59ba2c1ab04390a2592a4f971944c453257d4dfe2403918
3
+ size 7474559185
audio/eval_far_audio.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15319f3473a6285c0e58c97ec07c2204ae0771d618e2a237d0c4b14682d7734c
3
+ size 9117388635
audio/eval_near_audio.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cef343c1f0049746b926e0d4a183b7b77375965f6a44e2704a349ce50f382aa1
3
+ size 8092327904
audio/train_far_audio_00.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fb6df5f31a206d903db1f125c68015d69c2ea27aefb25280b093cec44770ac6
3
+ size 10737418240
audio/train_far_audio_01.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:160e3be4abfa0deef5668b601d5daae05fa65456f796bbb95a8bcc91f045e29e
3
+ size 10737418240
audio/train_far_audio_02.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:313cd0aa1c0faab5f04387a24821f4d253c9cc07d89b6bab6e6e5a13900f5976
3
+ size 10737418240
audio/train_far_audio_03.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60265159c7b50dc521373a4f0a7206a325c69cdb780bd8be6ef7270fddb6d210
3
+ size 10737418240
audio/train_far_audio_04.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be85c1c4195595dd4a111db68bf9a0a92754228647b2d777e186decec4f39eed
3
+ size 10737418240
audio/train_far_audio_05.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d597fd6e0d7438904cddbadb26ceebdaafce6c9f6b042a6be68a45cd0a64f8ca
3
+ size 9850534354
audio/train_near_audio_00.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:728a3ac6d9f3d90ad7303afa5d2bda9818a88d2287ab4f004054390de9bcf3dd
3
+ size 10737418240
audio/train_near_audio_01.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c27dbcac3c96889e438c5fe30c3e2a3b63566c6e053c741e05e11c6f04825473
3
+ size 10737418240
audio/train_near_audio_02.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9d602e26e4c77fab963dbd0a1d7838ffdbedf3dec8b860724a0571e3923643c3
3
+ size 10737418240
audio/train_near_audio_03.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a70ca8e5856d2dc9bcfb3a9c3cf9d680a91a45e52f246d6e8ebd3d66c5a1cea
3
+ size 10737418240
audio/train_near_audio_04.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:70f6166466253b247ade81846e7f072498e690462a69932f89ddfe7a268a01e7
3
+ size 8660881645
upload_aishell8_realscene_resume.log CHANGED
The diff for this file is too large to render. See raw diff
 
video/train_video_10.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b9b292e51db06665e44209bf68bc44cd0114f3fb246f33ad4fdd27450778ce5
3
+ size 10737418240