Mumospee V2: A MUltiMOdal SPEEch Corpus Version 2
Dataset Description
MuMoSPEE v2 is a large-scale multimodal speech dataset containing audio and video recordings with transcripts, aggregated from:
- European Council events — official speeches, interviews, and doorstep appearances.
- Public YouTube meetings — webinars, conferences, panel discussions, and institutional videos.
This release is built upon the first version of the MuMoSPEE dataset, significantly expanding both the scale and diversity of content, and unifying audio and video data into a consistent metadata format suitable for large-scale speech and multimodal research.
The original MuMoSPEE v1 dataset is available at:
👉 https://huggingface.co/datasets/meetween/mumospee
The dataset is designed for research in speech recognition, multimodal modeling, meeting analysis, and AI-driven content understanding.
All media are linked via URLs, with transcripts and metadata included. Audio and video recordings are stored as separate entries.
Sources & HuggingFace pages:
- EU Council: https://huggingface.co/datasets/meetween/eu_council. Original source at European Council Newsroom.
- YouTube Meetings: https://huggingface.co/datasets/meetween/meetween_youtube_meeting
Dataset Structure
| Column | Type | Description |
|---|---|---|
url |
string | URL to audio or video |
type |
string | audio or video |
duration |
float | Duration in seconds |
language |
string | Primary spoken language |
transcript |
string | Full transcript text |
tag |
string | Source tag (EU_Council or YouTube_Meeting) |
split |
string | Dataset split (train, validation, test) |
license |
string | License information for the content |
Statistic Summary
Audio (EU Council only)
- Number of audio entries: 45,522
- Total duration (hours): 3,673.86
- Average duration (seconds): 290.54
Video
- Number of video entries: 167,929
- Total duration (hours): 83,801.51
- Average duration (seconds): 1,796.51
Breakdown by tag:
| Tag | Count |
|---|---|
| EU_Council | 45,522 |
| YouTube_Meeting | 122,407 |
Language Distribution
| Language | Audio Count | Audio Hours | Video Count | Video Hours |
|---|---|---|---|---|
| English | 23,635 | 1,907.4 | 119,531 | 76,000 |
| Spanish | 2,427 | 196.7 | 13,810 | 8,785 |
| French | 2,953 | 238.6 | 11,176 | 7,110 |
| German | 3,582 | 289.2 | 5,049 | 3,210 |
| Portuguese | 1,230 | 99.5 | 4,260 | 2,710 |
| Italian | 1,374 | 111.2 | 2,845 | 1,810 |
| Dutch | 1,064 | 86.2 | 1,846 | 1,174 |
| Swedish | 1,092 | 88.5 | 1,174 | 746 |
| Polish | 970 | 78.7 | 999 | 635 |
| Czech | 873 | 70.9 | 886 | 563 |
| Croatian | 826 | 67.1 | 828 | 526 |
| Danish | 787 | 63.9 | 792 | 503 |
| Slovak | 684 | 55.5 | 685 | 435 |
| Finnish | 647 | 52.6 | 652 | 414 |
| Greek | 604 | 49.1 | 617 | 392 |
| Slovenian | 519 | 42.2 | 519 | 330 |
| Bulgarian | 401 | 32.6 | 403 | 256 |
| Hungarian | 384 | 31.2 | 386 | 245 |
| Luxembourgish | 297 | 24.1 | 297 | 188 |
| Romanian | 295 | 23.9 | 296 | 187 |
| Maltese | 156 | 12.6 | 156 | 99 |
| Multilingual | 94 | 7.6 | 94 | 60 |
| Lithuanian | 92 | 7.4 | 92 | 58 |
| Arabic | 79 | 6.5 | 79 | 50 |
| Latvian | 73 | 6.0 | 73 | 46 |
| Ukrainian | 56 | 4.7 | 56 | 35 |
| Russian | 38 | 3.2 | 38 | 24 |
| Estonian | 37 | 3.0 | 37 | 23 |
| Serbian | 36 | 2.9 | 36 | 22 |
| Georgian | 33 | 2.6 | 33 | 20 |
| Norwegian | 29 | 2.3 | 29 | 17 |
| Albanian | 25 | 2.0 | 25 | 15 |
| Macedonian | 23 | 1.9 | 23 | 14 |
| Bosnian | 17 | 1.4 | 17 | 10 |
| Montenegrin | 16 | 1.3 | 16 | 9 |
| Turkish | 14 | 1.1 | 14 | 8 |
| Belarusian | 6 | 0.5 | 6 | 3 |
| Moldavian | 5 | 0.4 | 5 | 3 |
| Japanese | 4 | 0.3 | 4 | 2 |
| Persian | 4 | 0.3 | 4 | 2 |
| Catalan | 4 | 0.3 | 4 | 2 |
| Chinese | 3 | 0.2 | 3 | 1 |
| Korean | 3 | 0.2 | 3 | 1 |
| Vietnamese | 3 | 0.2 | 3 | 1 |
| Armenian | 3 | 0.2 | 3 | 1 |
| Indonesian | 3 | 0.2 | 3 | 1 |
| Swahili | 1 | 0.1 | 1 | 0.1 |
| Hindi | 1 | 0.1 | 1 | 0.1 |
| Tajik | 1 | 0.1 | 1 | 0.1 |
| Kazakh | 1 | 0.1 | 1 | 0.1 |
| Khmer | 1 | 0.1 | 1 | 0.1 |
- Notes
For recordings containing multiple spoken languages, the total duration is split equally among the detected languages, as precise language-level timestamps are not available. This avoids inflating the total duration across languages.
Notes on Transcripts
EU Council (audio & video):
Transcripts are generated using the Whisper ASR package. Each entry contains one or more languages per recording, corresponding to the full speech.YouTube Meeting (video):
Transcripts are extracted from YouTube subtitle tracks. They may be auto-generated, and some videos may lack transcripts entirely.
Usage
from datasets import load_dataset
dataset = load_dataset("meetween/mumospee_v2")
# Access first audio sample
audio_sample = dataset['train'].filter(lambda x: x['type'] == 'audio')[0]
print(audio_sample['transcript'])
print(audio_sample['url'])
# Access first video sample
video_sample = dataset['train'].filter(lambda x: x['type'] == 'video')[0]
print(video_sample['transcript'])
print(video_sample['url'])
Licensing Information
- EU Council: Attribution only, non-commercial (© EU 2025). The copyright notice of the dataset: https://www.consilium.europa.eu/en/about-site/copyright/
- YouTube: Metadata and transcripts are CC0 / Public Domain; videos follow original uploader’s license (CC-BY).
Users must comply with the source license when accessing media via URLs.
- Downloads last month
- 10