Parliaments, councils, courts, police and public services use speech recognition to produce records of meetings, caption public sessions, and transcribe interviews and citizen calls. The technical problems are familiar: long meetings, many speakers, local names. What sets the public sector apart is its obligations: accessibility law, open government and archiving rules, a public-task legal basis under the GDPR, procurement rules, and growing expectations of data sovereignty.
Key takeaways
- Typical uses are meeting minutes, parliamentary reports, captions on webcast sessions, citizen contact centres, and interviews in policing, social services and immigration.
- Official records are edited, not raw. ASR produces the draft; reporters and clerks produce the record.
- The Web Accessibility Directive requires public bodies to make their online content accessible, including captions for pre-recorded video, with an exemption for live streams.
- Transcripts can become public records, subject to freedom-of-information and archiving laws. Decide in advance what is kept and for how long.
- Regional and minority languages are a real gap in most commercial systems. Check coverage before procurement.
Where speech recognition is used#
| Setting | Output | Main difficulty |
|---|---|---|
| Parliaments and assemblies | Verbatim reports, captions, search | Many speakers, interruptions, procedure vocabulary |
| Municipal councils | Minutes, webcast captions | Room microphones, local place names, dialect |
| Citizen contact centres | Call notes, analytics, voice agents | Telephony audio, many languages |
| Police interviews | Interview transcripts for case files | Evidence standard, interpreters |
| Social services and health | Casework notes | Sensitive data, informal speech |
| Immigration and asylum interviews | Interview records | Interpreted speech, high stakes |
Courts and tribunals are covered in legal transcription, and citizen contact centres in contact centre speech analytics.
Parliaments and councils#
Parliamentary records are edited transcripts, not literal ones. The UK's Hansard describes itself as a substantially verbatim report: repetitions and obvious mistakes are removed, meaning is kept. The European Parliament publishes verbatim reports of plenary debates, with each speech in the language in which it was delivered. Many national parliaments and larger councils now use ASR to produce a first draft, which parliamentary reporters then correct and edit to house style.
The acoustic setting helps: each seat usually has its own microphone and the chair controls who speaks. It also creates problems: several microphones open during interruptions, members speaking away from the microphone, and heckling. Speaker identification is usually solved from the conference system, which knows which microphone is live, rather than from diarisation.
Council meetings are harder. Smaller rooms have fewer microphones, members of the public speak, and discussion refers to local streets, projects and people. Supplying the agenda, the list of councillors and planning documents as context covers most of the proper nouns.
Accessibility#
The Web Accessibility Directive (Directive (EU) 2016/2102) requires public sector bodies to make their websites and mobile apps accessible, with EN 301 549 as the harmonised standard, which in turn draws on WCAG. For video, this means captions on pre-recorded content. The directive excludes live time-based media from its scope, and pre-recorded time-based media published before 23 September 2020, but many public bodies caption live council and parliament webcasts anyway and publish captioned recordings afterwards, which do fall under the requirements.
Captioning a four-hour council meeting by hand is expensive; ASR makes it routine. The quality bar for accessibility is readability and correct names, so the workflow issues in media and broadcast transcription and captions and subtitles apply directly.
Open government and records#
Once created, a transcript held by a public body is generally a record. It may have to be disclosed under freedom-of-information laws, such as the UK's Freedom of Information Act 2000 or the Dutch Open Government Act (Wet open overheid), and kept or destroyed according to archiving rules. Two consequences follow:
- Decide what the record is. If the official minutes are the record, a raw ASR transcript may be a working document to delete once minutes are approved. If the transcript is published, it should be corrected first.
- Errors become public. A misrecognised name in a published transcript can misattribute a statement to a councillor or citizen. Review before publication, especially for names and votes.
Legal basis, sensitive processing and procurement#
Data protection#
Public bodies usually process personal data in recordings under Article 6(1)(e) of the GDPR, performance of a task in the public interest or in the exercise of official authority, with the task laid down in law. Social services and health recordings add special category data under Article 9. Police processing for law enforcement purposes falls under the Law Enforcement Directive (Directive (EU) 2016/680) instead of the GDPR. A speech recognition vendor acts as a processor and needs an Article 28 agreement; see data processing agreements. A data protection impact assessment is likely required for sensitive or large-scale uses.
AI Act#
The EU AI Act lists several public-sector uses as high-risk in Annex III, including AI systems used to evaluate eligibility for public assistance benefits, and systems used in migration, asylum and border control management. Transcription on its own is generally a supporting tool rather than one of these systems, but a pipeline that transcribes interviews and then scores credibility or eligibility is a different matter. See the EU AI Act and speech recognition.
Procurement and sovereignty#
Contracts above EU thresholds are tendered under Directive 2014/24/EU, and tender documents increasingly specify processing location, certifications and exit terms. Many public bodies require processing within the EU, or deployment in their own data centre or government cloud. See EU data residency, on-premise and VPC deployment and sovereign AI in Europe.
Official, regional and minority languages#
Public bodies often have legal duties to work in more than one language: Welsh in Wales, Frisian in Friesland, Catalan, Basque and Galician in Spain, Swedish in Finland. Most commercial speech recognition covers the large state languages well and regional languages poorly or not at all. Citizens also speak to public services in their own languages, which pushes contact centres towards multilingual recognition and language identification. Check language coverage against your statutory duties before procurement, and test on local speakers and dialects, not only on standard broadcast speech.
How Reson8 approaches this#
Resonant-1 covers nine European languages in depth: Dutch, English, French, German, Italian, Polish, Portuguese, Spanish and Swedish. It does not cover regional languages such as Frisian, Welsh or Catalan, so a public body with duties in those languages would need another solution for them. Processing runs on EU GPUs with zero audio retention and no training on customer data, under a GDPR Article 28 DPA; Resonate Labs is ISO 27001 certified, and on-premise and VPC deployment are available on enterprise plans. VoxPopuli was excluded from Reson8's reported benchmark averages because it was used in training.
Common questions#
Can ASR replace parliamentary reporters?
It changes their work from typing to editing. Official reports follow editorial conventions, and someone accountable has to decide what the record says.
Do live council webcasts need captions?
The Web Accessibility Directive excludes live time-based media, but national law or policy may go further, and the recording published afterwards is in scope. Captioning live is good practice where feasible.
Can transcripts be requested under freedom of information?
Usually, if the body holds them, subject to the normal exemptions. Retention decisions made in advance avoid holding transcripts nobody needs.
What about interpreted interviews?
Transcribe both the original speaker and the interpreter, and label them. Errors in either can matter, and the record should show what each party actually said.
References#
- Directive (EU) 2016/2102 on the accessibility of the websites and mobile applications of public sector bodies. EUR-Lex.
- ETSI. EN 301 549: Accessibility requirements for ICT products and services.
- Wang, C. et al. (2021). VoxPopuli: A Large-Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretation. ACL. arXiv:2101.00390.
- Regulation (EU) 2016/679 (General Data Protection Regulation), Article 6(1)(e). EUR-Lex.
- Directive (EU) 2016/680 (Law Enforcement Directive). EUR-Lex.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Annex III. EUR-Lex.
- Directive 2014/24/EU on public procurement. EUR-Lex.