Datasets:

meetween
/

eu_council

size_categories:
  - 10K<n<100K
pretty_name: Meetween EU Council Newsroom Dataset
dataset_info:
  features:
    - name: id
      dtype: string
    - name: title
      dtype: string
    - name: url
      struct:
        - name: audio
          dtype: string
        - name: video
          dtype: string
        - name: video_preview
          dtype: string
    - name: date
      dtype: string
    - name: duration
      dtype: float64
    - name: language
      list: string
    - name: transcript
      dtype: string
    - name: segments
      list:
        - name: end
          dtype: float64
        - name: language
          dtype: string
        - name: speaker
          dtype: string
        - name: start
          dtype: float64
        - name: text
          dtype: string
    - name: word_count
      dtype: int64
    - name: character_count
      dtype: int64
    - name: videoTypeDescription
      dtype: string
    - name: split
      dtype: string
    - name: license
      dtype: string
    - name: attribution
      struct:
        - name: published_at
          dtype: string
        - name: publisher
          dtype: string
        - name: ref
          dtype: string
        - name: title
          dtype: string
  splits:
    - name: eu_council_2
      num_bytes: 481959645
      num_examples: 45411
  download_size: 239625086
  dataset_size: 481959645
configs:
  - config_name: default
    data_files:
      - split: eu_council_2
        path: data/eu_council_2-*

Meetween EU Council Newsroom Dataset

Dataset Description

This dataset contains transcripts, audio, and video of official “Arrival” and “Doorstep” speeches from European Council events, collected from the European Council Newsroom. Each entry includes metadata such as the title, date, duration, language, word/character counts, and links to audio and video sources. The speeches are mainly in European languages, with some multilingual and non-European entries.

All recordings were automatically transcribed using OpenAI’s Whisper ASR model. Items lacking audio or containing two audio channels (≈6% of recordings) were excluded to simplify processing.

Besides, items whose language label are "Multilingual" or "International" are exccluded.

Source: European Council Newsroom

Dataset Structure

Column Name	Data Type	Description
`title`	string	Title of the entry
`date`	string / date	Uploaded date on the webpage
`duration`	float	Duration in seconds
`language`	list of string	Languages spoken in speech
`transcript`	string	Full text transcript of the speech
`segments`	list[dict]	a time-aligned transcription of the media content.
`word_count`	int	Number of words in the transcript
`character_count`	int	Number of characters in the transcript
`videoTypeDescription`	string	Type of video (e.g. doorstep interview, speech)
`split`	string	Data split
`license`	string	Dataset license information
`url.audio`	string	URL to the audio
`url.video_preview`	string	URL to the preview of the video
`url.video`	string	URL to the HD video

Dataset Size

Number of records: 45411
Total duration in hours: 3665.68
Average transcript length: 3951.56 characters / 660.20 words

Language Distribution

Language	Count	Total Duration (hours)
English	23635	1754.21
German	3582	294.16
French	2953	240.68
Spanish	2427	230.31
Italian	1374	122.0
Portuguese	1230	113.16
Swedish	1092	96.51
Dutch	1064	97.09
Polish	970	91.22
Czech	873	77.6
Croatian	826	71.11
Danish	787	71.06
Slovak	684	62.17
Finnish	647	62.9
Greek	604	37.16
Slovenian	519	40.49
Bulgarian	401	66.11
Hungarian	384	34.39
Luxembourgish	297	28.92
Romanian	295	25.66
Maltese	156	5.63
Lithuanian	92	5.32
Arabic	79	4.53
Latvian	73	5.18
Ukrainian	56	4.48
Russian	38	2.67
Estonian	37	2.36
Serbian	36	2.84
Georgian	33	1.99
Norwegian	29	2.98
Albanian	25	1.89
Macedonian	23	1.78
Bosnian	17	1.36
Montenegrin	16	1.09
Turkish	14	1.37
Belarusian	6	0.31
Moldavian	5	0.34
Catalan	4	0.41
Persian	4	0.36
Japanese	4	0.25
Indonesian	3	0.25
Viêt Namese	3	0.33
Chinese	3	0.28
Armenian	3	0.07
Korean	3	0.33
Hindi	1	0.1
Kazakh	1	0.1
Khmer	1	0.06
Tajik	1	0.09
Swahili	1	0.05

Note:
For items 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.

Usage

from datasets import load_dataset

# Load the full dataset. 
ds = load_dataset("meetween/eu_council", split="eu_council")

# Load dataset in streaming mode.
dataset = load_dataset(
    "meetween/eu_council",
    split="eu_council",
    streaming=True
)
for example in dataset.take(3):
    print(example["id"], example["language"])