About
Automatic transcripts for UN Web TV meetings.
What is this?
UN Web TV Transcripts is a public preview tool that automatically generates searchable text transcripts from United Nations meeting recordings published on UN Web TV (opens in new tab).
Who is it for?
The tool is designed to help anyone who needs quick access to the spoken content of UN meetings, including:
- Diplomats and delegation staff following proceedings across multiple organs
- Researchers and academics studying UN debates and voting records
- Journalists covering United Nations affairs
- Civil society organisations monitoring policy discussions
- UN Secretariat staff reviewing meeting records
What meetings are covered?
The tool covers public meetings recorded and published on UN Web TV, including:
- Security Council (SC) — open meetings and briefings
- General Assembly (GA) — plenary and main committee sessions
- Human Rights Council (HRC)
- Economic and Social Council (ECOSOC)
- Other inter-governmental bodies as available on Web TV
Closed or confidential meetings are not recorded on Web TV and are therefore not covered.
How it works
1
Meeting schedule collection
The system monitors UN Web TV for newly published meeting recordings. Meeting metadata — title, date, UN organ, and document references — is extracted from the Web TV website and stored.
2
Audio transcription
We transcribe the existing UN audio channels — the original "floor" mix and each official UN interpretation channel — directly. No machine translation is applied. Each channel is transcribed by a speech-to-text model chosen for that language.
3
Speaker identification
After transcription, a second AI model analyses the text and audio to assign names and affiliations to each speaker where possible. It uses contextual clues — such as the chair introducing delegates, country name mentions, and speaker diarization — together with the official list of participants when available.
4
Topic analysis
The transcript is automatically analysed to identify the main policy topics discussed, using categories relevant to UN proceedings (e.g. humanitarian affairs, international peace and security, human rights).
5
Official record alignment
Where official verbatim or summary records (PV/SR documents) exist, the system retrieves them and adds timestamps to align the text with the video. Besides timestamps, no AI processing is involved for these documents.
Accuracy and limitations
These transcripts are created using automatic speech recognition and are not official UN records.
They are a faster, unofficial reference that may be useful for monitoring and research, but should not be cited as authoritative. For the official record, please refer to the UN Official Document System (opens in new tab) (verbatim records, summary records, and resolutions).
Automatic transcription is much faster than human transcription but introduces errors a human reviewer would catch. Common issues include:
- Proper nouns — country names, delegate names, place names, and UN document symbols may be misheard or misspelt
- Technical terminology — legal or procedural phrases specific to UN practice may be transcribed incorrectly
- Accented speech — accuracy varies by speaker accent and microphone quality
- Overlapping speech — when multiple speakers talk simultaneously, attribution may be wrong
Data sources
- UN Web TV — Meeting recordings and metadata, delivered via the Kaltura media platform (publicly accessible at webtv.un.org).
- UN Official Document System — Official verbatim records (PV documents) retrieved from documents.un.org where available.
Status
This tool is in Public Preview. Features, coverage, and accuracy are actively being improved. Feedback is welcome.