Speech in the Browser for UK Services: en-GB, Dragon, GDPR
Set the Web Speech API to British English, test speech features the GOV.UK way with Dragon, and know where staff audio goes before you build dictation in.
By Redbison Dev · · 7 min read

United Kingdom edition. Written for UK developers and users: en-GB recognition and voices, the public sector accessibility regulations and GOV.UK service manual testing with Dragon, Access to Work, and what browser dictation means under UK GDPR. US edition
If you build or use speech features in a browser in the UK, three local details change how the Web Speech API should be used. Recognition should be told to listen for British English with lang = 'en-GB', because the browser's speech service picks its model from that tag. Public sector services must work for people who already use speech recognition software such as Dragon, which is a different thing from a page that listens through the Web Speech API. And in Chrome the audio a page records for recognition goes to Google's servers, which matters when the speaker is a member of staff or a customer.
This edition covers each with code, the GOV.UK testing guidance, and the British English version of Speech to Text, which starts in English (UK). For how the API works underneath, see the parent guide.
Ask for British English, in both directions
Recognition. The lang property tells the recogniser which language model to use. MDN says that if it is not set, it defaults to the HTML lang attribute, or the browser's language setting if that is missing too, so a UK site with <html lang="en-GB"> gets British English by default. Setting it explicitly is safer when the page language is missing or generic:
const Rec = window.SpeechRecognition || window.webkitSpeechRecognition;
const rec = new Rec();
rec.lang = document.documentElement.lang || 'en-GB';
Our British English page does this for you: its language picker starts at English (UK) instead of English (US), and lists Irish, Indian, Australian, New Zealand, Singapore, Philippine and Pakistani English as alternatives for users who speak those varieties.
Speech synthesis. Voice lists vary by device, so pick by language tag, falling back from an exact en-GB match to any English voice:
function ukVoice() {
const voices = speechSynthesis.getVoices();
return voices.find(v => v.lang === 'en-GB' && v.localService)
?? voices.find(v => v.lang === 'en-GB')
?? voices.find(v => v.lang.startsWith('en'))
?? null;
}
The first line prefers a local voice, because a voice with localService set to false sends your text to a server to be synthesised. On a council or NHS page reading back a form, that preference keeps the user's answers on their own device. Remember that the list often arrives late, after a voiceschanged event, so call ukVoice() again when it fires.
GOV.UK's guidance on WCAG 2.2 also asks services to make clear what language the content is written in. Setting lang="en-GB" on the page serves both purposes: screen readers and synthesis voices pronounce it correctly, and recognition defaults to the right model.
Public sector duties: work with the user's own speech software
The Public Sector Bodies (Websites and Mobile Applications) (No. 2) Accessibility Regulations 2018 came into force on 23 September 2018. GOV.UK's guidance says public sector websites and apps must meet WCAG 2.2 AA (unless there are valid legal reasons not to) and publish an accessibility statement, and that the regulations build on duties under the Equality Act 2010, or the Disability Discrimination Act 1995 in Northern Ireland.
Most people who control a computer by voice do not use a web page's microphone button. They use desktop speech recognition such as Dragon, which turns their speech into clicks and typing on any page. GOV.UK's service manual lists Dragon 15 or later with Chrome among the combinations to test, and asks testers to use speech to:
- navigate to each feature;
- activate every link, button and interactive element, such as form controls or a media player;
- enter text into any forms.
It also advises a good headset rather than a built-in microphone, speaking clearly but naturally, saying punctuation aloud and correcting mistakes as you go.
Worked example: a button a voice user cannot press
A form has a button that reads Continue on screen, but a developer has given it aria-label="Submit postcode lookup". A sighted Dragon user says "click Continue", the word they can see, but the button's accessible name is something else, so the command may not find it. GOV.UK's WCAG 2.2 overview asks for accessible labels that match or closely resemble the visible label, and this is why.
The fix is to drop the aria-label or make it start with the visible text:
<!-- Fails for voice users -->
<button aria-label="Submit postcode lookup">Continue</button>
<!-- Works: the accessible name is the visible text -->
<button>Continue</button>
Run the service manual's three tests (navigate, activate, enter text) on every page with Dragon, and treat any element that cannot be activated by its visible name as a defect.
Adding your own speech features
If you add a dictation button to a service, make it an extra, never the only way in:
- Firefox has no SpeechRecognition interface, so a dictation button must fall back to a normal text field.
- Recognition needs a network connection in Chrome and Edge, and fails offline with a
networkerror. - Do not let it conflict with Dragon. A user running desktop speech software does not need the page to open the microphone as well.
Help with dictation software at work: Access to Work
For someone who dictates because of a disability or health condition, the UK has practical funding. Access to Work can give a grant to help pay for specialist equipment and assistive software, as well as support workers. GOV.UK says the grant does not affect other benefits and does not have to be paid back. Northern Ireland runs its own scheme, Access to Work (NI), through nidirect and the Department for Communities, which can help with special aids and equipment. Employers and employees can point staff to it before buying dictation software, and IT teams should check that approved speech software works with their internal systems, which public sector intranets are also expected to make accessible.
Where the audio goes, and UK GDPR
In Chrome the microphone stream is sent to Google's speech service and the text comes back over the network. The page never sees the audio, but the browser vendor does.
UK GDPR defines personal data as any information relating to an identified or identifiable person. A recording of a named employee or customer speaking, and the transcript of what they said, will usually be personal data. So before you add browser dictation to an internal tool or a customer-facing form:
- Tell people that their speech is sent to an outside service for recognition.
- Offer a typed alternative that sends nothing.
- Keep it away from sensitive content, such as health details, case notes or anything about third parties, unless your data protection officer has approved that use.
- Prefer local synthesis voices for reading back what users typed.
For individual users, the Speech to Text page says the same thing in plain terms: dictation in Chrome is not private, so keep passwords, patient details and confidential material out of it.
Tools for this job
- Speech to Text (British English): dictation that starts in English (UK), with other regional English options and an optional recording.
- Text to Speech: hear text in a British English voice, with online voices labelled so you can keep text on your device.
- Audio Format Converter: convert a recording when another program needs WAV.
Frequently asked questions
How do I make the Web Speech API recognise British English?
Set recognition.lang = 'en-GB', or make sure the page has lang="en-GB" on its html element, which the recogniser uses when lang is not set. On our British English Speech to Text page, the language already starts as English (UK).
How do I pick a British voice for speechSynthesis?
Filter speechSynthesis.getVoices() for lang === 'en-GB', prefer voices with localService set to true, and fall back to any English voice. Call it again on voiceschanged, because the list often loads after the page.
Do public sector websites have to work with speech recognition?
Public sector websites and apps must meet WCAG 2.2 AA under the accessibility regulations, and GOV.UK's service manual says services must work with assistive technologies including speech recognition software, listing Dragon with Chrome among the combinations to test.
Is browser dictation allowed at work under UK GDPR?
The law does not ban it, but a recording or transcript of an identifiable person is usually personal data, and in Chrome the audio goes to Google's servers. Tell people, offer a typed alternative, and check with your data protection officer before using it for anything sensitive.
Sources and further reading
- Understanding accessibility requirements for public sector bodies, GOV.UK. In force since 23 September 2018; WCAG 2.2 AA; accessibility statements; Equality Act 2010.
- Testing with assistive technologies, GOV.UK Service Manual. Dragon 15 or later with Chrome; navigate, activate and enter text by voice.
- Understanding WCAG 2.2, GOV.UK Service Manual. Labels that match the visible label; making the content language clear.
- Access to Work, GOV.UK. Grants for specialist equipment and assistive software.
- Access to Work (NI): practical help at work, nidirect. The Northern Ireland scheme, run by the Department for Communities.
- UK GDPR Article 4: definitions, legislation.gov.uk. Article 4(1), personal data.
- SpeechRecognition: lang property, MDN Web Docs. Defaults to the HTML lang attribute.
- SpeechSynthesisVoice: localService property, MDN Web Docs.
More Text/Speech Tools
Every tool in the Text/Speech Tools section, free and in your browser.