EuroSpeech Data Sources

This repository contains the source metadata for the EuroSpeech multilingual speech dataset. EuroSpeech is a large-scale parliamentary speech corpus covering 22 European languages with over 78k hours of aligned speech-text data.

📋 Repository Contents

This repository provides comprehensive CSV files that document the original sources for all audio and transcript data used in the EuroSpeech dataset. For each of the 22 countries included in EuroSpeech, we maintain detailed metadata files containing:

Video/Audio URLs: Direct links to parliamentary recordings
Transcript URLs: Links to corresponding transcript documents
Session metadata: Parliamentary session identifiers and additional context

🔗 Relationship to EuroSpeech Dataset

The main EuroSpeech dataset is available at: disco-eth/EuroSpeech

Each audio segment in the main EuroSpeech dataset contains video_id and/ or transcript_id fields that can be cross-referenced with the CSV files in this repository to trace back to the original parliamentary source URLs. This provides complete transparency and enables researchers to:

Verify the provenance of specific data points
Understand the original context of parliamentary sessions
Replicate or extend the data collection methodology

📁 File Structure

The CSV files vary significantly in structure across different parliaments. Each country's CSV reflects the specific data available from their parliamentary systems:

Core Identifier Fields:

video_id - Unique identifier for video/audio (always present)
transcript_id - Unique identifier for transcript (when available; otherwise video_id serves for both)

URL Fields (examples from actual implementation):

Video/Audio Sources:

mp4_video_link, youtube_link, m3u8_link, mp3_link
generic_video_link, generic_m3u8_link, processed_video_link

Transcript Sources:

pdf_link, html_link, dynamic_html_link, doc_link
processed_transcript_html_link, processed_transcript_text_link

Subtitle Sources:

srt_link

Metadata Fields:

Highly variable across parliaments
May include dates, session information, titles, etc.
Structure and availability depend entirely on what each parliament provides

Example CSV structure:

video_id,mp4_video_link,doc_link,processed_transcript_html_link
12345,https://example.com/video/12345.mp4,https://example.com/transcript/12345.docx,https://example.com/transcript/12345

Note: While the examples above show the supported column types from our pipeline, the actual column names and available fields vary by parliament, even thought the url fields should always be a subset of the provided ones.

🌍 Countries and Languages Covered

The repository includes source metadata for parliamentary data from 22 European countries:

🇭🇷 Croatia (hr) - 🇩🇰 Denmark (da) - 🇳🇴 Norway (no) - 🇵🇹 Portugal (pt)
🇮🇹 Italy (it) - 🇱🇹 Lithuania (lt) - 🇬🇧 United Kingdom (en) - 🇸🇰 Slovakia (sk)
🇬🇷 Greece (el) - 🇸🇪 Sweden (sv) - 🇫🇷 France (fr) - 🇧🇬 Bulgaria (bg)
🇩🇪 Germany (de) - 🇷🇸 Serbia (sr) - 🇫🇮 Finland (fi) - 🇱🇻 Latvia (lv)
🇺🇦 Ukraine (uk) - 🇸🇮 Slovenia (sl) - 🇪🇪 Estonia (et) - 🇧🇦 Bosnia & Herzegovina (bs)
🇮🇸 Iceland (is) - 🇲🇹 Malta (mt)

🔍 Data Sources

Parliamentary data was primarily sourced from:

Official parliament websites of each respective country
YouTube channels operated by parliamentary institutions
Dedicated parliamentary media portals and archives

All sources consist of publicly available parliamentary recordings and official transcripts as provided by the respective governmental institutions.

⚖️ Copyright and Legal Information

Detailed copyright and licensing references for each country's parliamentary data is available in the main EuroSpeech paper.

🔬 Research Applications

This metadata repository enables several research applications:

Reproducibility: Complete replication of the EuroSpeech data collection process
Extension: Adding new parliamentary sources or time periods to the dataset
Analysis: Studying characteristics and biases of different parliamentary sources
Verification: Cross-checking data quality and source authenticity
Methodology: Understanding the data collection pipeline for similar projects

📚 Citation

If you use this source metadata in your research, please cite the EuroSpeech paper:

@article{eurospeech2025,
  title={EuroSpeech: A Multilingual Speech Corpus},
  author={Pfisterer, Samuel and Grötschla, Florian and Lanzendörfer, Luca and Yan, Florian and Wattenhofer, Roger},
  journal={TBD},
  year={2025}
}

🤝 Contributing

We welcome contributions to expand this metadata collection. If you have additional parliamentary sources or corrections to existing metadata, please open an issue or submit a pull request.

📞 Contact

For questions about this repository or the EuroSpeech dataset:

GitHub: EuroSpeech Repository
Email: spfisterer@ethz.ch

This repository is part of the EuroSpeech project conducted at ETH Zurich's DISCO Group.