text stringlengths 0 1.08k |
|---|
Raw-video eval clips (H.264 mp4, CFR at source fps, no audio) + same-stem .txt transcripts. |
Originals: ../raw_src (fetched by ../_work/fetch_media.py; normalised by ../_work/normalize.py). |
Per-clip source URL / license: sources.json. Do not commit or redistribute (gitignored). |
RAVDESS - CC BY-NC-SA 4.0 |
Livingstone & Russo (2018), The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), PLoS ONE 13(5): e0196391; zenodo.org/records/1188976 |
clips: ravdess_a01_s1, ravdess_a02_s2, ravdess_a07_s2, ravdess_a10_s1, ravdess_a13_s1, ravdess_a18_s2 |
CREMA-D - ODbL 1.0 (database) + DbCL 1.0 (contents) |
Cao et al. (2014), CREMA-D: Crowd-sourced Emotional Multimodal Actors Dataset, IEEE Trans. Affective Computing 5(4); github.com/CheyneyComputerScience/CREMA-D |
clips: cremad_1001_mti, cremad_1007_iom, cremad_1013_iwl, cremad_1019_tie, cremad_1024_tai, cremad_1032_wsi, cremad_1037_ith, cremad_1034_iww |
GRID - CC BY 4.0 |
Cooke, Barker, Cunningham & Shao (2006), An audio-visual corpus for speech perception and automatic speech recognition, JASA 120(5); zenodo.org/records/3625687 |
clips: grid_s1_bbaf2n, grid_s1_pbav2n, grid_s4_bbae9n, grid_s4_pbao6s, grid_s2_bbaf1n, grid_s2_pbapza |
Maybe tomorrow it will be cold |
I'm on my way to the meeting |
I would like a new alarm clock |
That is exactly what happened |
The airplane is almost full |
We'll stop in a couple of minutes |
I wonder what this is about |
I think I have a doctor's appointment |
bin blue at f two now |
place blue at v two now |
bin blue at f one now |
place blue at p zero again |
bin blue at e nine now |
place blue at o six soon |
Kids are talking by the door |
Dogs are sitting by the door |
Dogs are sitting by the door |
Kids are talking by the door |
Kids are talking by the door |
Dogs are sitting by the door |
raw_eval_v2: real-face eval clips (H.264 mp4, CFR at source fps, no audio) + same-stem .txt |
transcripts. Built by ml/scripts/build_eval_v2.py; per-clip source, licence and attributes in |
manifest.json. Private eval only: do not commit or redistribute; respect each licence below. |
None of these speakers is in the model's training data (LRS2, LRS3, VoxCeleb2, AVSpeech (Auto-AVSR, Ma et al. 2023: 3,448 h)). |
CREMA-D - ODbL 1.0 (database) + DbCL 1.0 (contents) - top level |
Cao et al. (2014), CREMA-D: Crowd-sourced Emotional Multimodal Actors Dataset, IEEE Trans. Affective Computing 5(4); https://github.com/CheyneyComputerScience/CREMA-D |
60 clips, 30 speakers: cremad_1005_iww, cremad_1005_tai, cremad_1012_ith, cremad_1012_mti, cremad_1014_iww, cremad_1014_tai, cremad_1015_ieo, cremad_1015_iom, cremad_1016_iwl, cremad_1016_tie, cremad_1020_ieo, cremad_1020_iom, cremad_1021_its, cremad_1021_wsi, cremad_1028_dfa, cremad_1028_tsi, cremad_1029_dfa, cremad... |
MEAD - data licence not stated (code MIT); research use only - mead/ (open (public Google Drive release; data licence not stated); no data licence published; used for private research evaluation only; bytes from the unofficial HF copy jordantencent/my_MEAD of the public Google Drive release) |
Wang et al. (2020), MEAD: A Large-scale Audio-visual Dataset for Emotional Talking-face Generation, ECCV; https://wywu.github.io/projects/MEAD/MEAD.html |
48 clips, 8 speakers: mead_m005_003_front, mead_m005_003_right_30, mead_m005_003_top, mead_m005_015_down, mead_m005_015_front, mead_m005_015_left_30, mead_m009_022_front, mead_m009_022_left_30, mead_m009_022_right_60, mead_m009_036_front, mead_m009_036_left_60, mead_m009_036_right_30, mead_m011_006_front, mead_m011_0... |
RAVDESS - CC BY-NC-SA 4.0 - top level |
Livingstone & Russo (2018), The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), PLoS ONE 13(5): e0196391; https://zenodo.org/records/1188976 |
8 clips, 8 speakers: ravdess_a04_s1, ravdess_a06_s1, ravdess_a11_s2, ravdess_a12_s2, ravdess_a14_s2, ravdess_a15_s1, ravdess_a21_s1, ravdess_a23_s2 |
VidTIMIT - CC BY-NC-ND 4.0 (Zenodo); cite Sanderson & Lovell (2009) - top level |
C. Sanderson and B.C. Lovell, Multi-Region Probabilistic Histograms for Robust and Scalable Identity Inference, LNCS 5558, 2009; https://zenodo.org/records/158963 |
28 clips, 16 speakers: vidtimit_fadg0_sa1, vidtimit_fadg0_sa2, vidtimit_faks0_sa1, vidtimit_faks0_sa2, vidtimit_fcft0_sa2, vidtimit_fedw0_sa1, vidtimit_fedw0_sa2, vidtimit_fjem0_sa1, vidtimit_fjem0_sa2, vidtimit_fjre0_sa1, vidtimit_fjre0_sa2, vidtimit_fjwb0_sa1, vidtimit_fjwb0_sa2, vidtimit_fpkt0_sa1, vidtimit_fpkt0_... |
I wonder what this is about |
The airplane is almost full |
I think I have a doctor's appointment |
Maybe tomorrow it will be cold |
I wonder what this is about |
The airplane is almost full |
It's eleven o'clock |
I'm on my way to the meeting |
I would like a new alarm clock |
That is exactly what happened |
It's eleven o'clock |
I'm on my way to the meeting |
I think I've seen this before |
We'll stop in a couple of minutes |
Don't forget a jacket |
The surface is slick |
Don't forget a jacket |
The surface is slick |
It's eleven o'clock |
I'm on my way to the meeting |
I think I've seen this before |
We'll stop in a couple of minutes |
I think I've seen this before |
We'll stop in a couple of minutes |
Don't forget a jacket |
The surface is slick |
I would like a new alarm clock |
That is exactly what happened |
I think I have a doctor's appointment |
Maybe tomorrow it will be cold |
I think I have a doctor's appointment |
Maybe tomorrow it will be cold |
I'm on my way to the meeting |
I would like a new alarm clock |
Don't forget a jacket |
The surface is slick |
I think I've seen this before |
We'll stop in a couple of minutes |
It's eleven o'clock |
I'm on my way to the meeting |
I would like a new alarm clock |
That is exactly what happened |
I would like a new alarm clock |
That is exactly what happened |
I wonder what this is about |
End of preview. Expand in Data Studio
No dataset card yet
- Downloads last month
- 140