AI Voice and Transcription Tools at Work
Share
Transcription is one of the few AI features that is genuinely close to solved, which makes the remaining failures easy to miss.
Direct answer
Modern transcription is accurate on clear single-speaker audio and degrades predictably with crosstalk, phone audio, jargon and some accents. Consent matters more than accuracy in most workplaces, and it is the part most often skipped.
Based on the AI and Automation Practitioner standard (ST1512), a Level 4 apprenticeship for people already in work. Funding rules described here are the England position and are set by the funding body.
Potentially free for you
The learner never pays. Employers potentially pay nothing either, and where the apprentice is aged 16 to 24 the government funds the whole cost even if the employer does not pay the levy.
Closing any gap. A levy transfer from another employer covers 100% of the cost. Unused transfer allowance is not carried forward at the end of the tax year, and it must be agreed before the apprentice starts.
From 1 August 2026 the government stopped adding the 10% top-up to new funds entering levy accounts, so there is less in the system than there was.
Part of our guide to AI Tools Worth Learning for Office Work.
How accurate it actually is
On clear single-speaker recording, error rates are low enough that the transcript is usable with light editing.
Crosstalk is the main failure. Overlapping speech produces merged or dropped text with no indication that anything was lost.
Phone and bridge audio is materially worse. Compression removes the frequencies these systems rely on.
Specialist vocabulary fails predictably. Product names, drug names, legal terms and surnames all get replaced by common words that sound similar.
Accents and fairness
Accuracy varies across UK regional accents. This is a documented limitation rather than an impression, and it matters.
It also varies for speakers whose first language is not English, and for some speech differences.
That has a practical consequence. If transcripts feed into a record that affects someone, uneven accuracy is a fairness problem.
Where transcripts are used in HR, disciplinary or clinical contexts, this needs an explicit human check rather than trust.
Consent and records
Tell people before recording. A notification banner is not the same as saying it out loud, particularly with external participants.
Decide retention. A permanent archive of every conversation is a liability nobody chose to create.
Know where the audio is stored and whether it leaves your organisation. Many tools process externally by default.
For anything involving personal data or employment matters, check whether a recording should exist at all before worrying about accuracy.
There is more on this in our guide to AI tools worth learning for office work.
Where it genuinely saves time
Interviews and research calls, where you need the exact words and would otherwise be typing rather than listening.
Dictating a first draft while walking or driving, which suits people who think better out loud.
Accessibility. Captions and transcripts make content usable for people who cannot use audio, which is a requirement in many settings.
Multilingual working, where a transcript can be translated far more cheaply than audio can be re-recorded.
Synthetic voice
Acceptable for internal announcements, training narration and accessibility features where no individual is implied.
Not acceptable for anything presented as a named person speaking, without their explicit consent in writing.
Cloning a real voice requires consent and usually a licence. Treat it like using someone's photograph rather than like using a font.
Disclose it in customer-facing content. The reputational cost of being found out exceeds the cost of a line of text.
Recording and consent in the UK
Recording a conversation and processing the audio is personal data processing under UK GDPR, and it needs a lawful basis.
For internal meetings that is usually straightforward. For calls with customers or the public it deserves an explicit decision rather than a default setting.
Tell people it is recording, out loud. A banner in the interface is not the same as informing someone, and external participants notice the difference.
Decide retention explicitly. An indefinite archive of every recorded conversation is a liability that nobody chose to create.
Choosing a tool
Check where audio is processed and stored, and for how long. This decides more than accuracy for most organisations.
Check whether it supports a custom vocabulary. Adding your product names and surnames removes the commonest error class.
Check speaker separation quality if you record meetings. That is where usable and unusable diverge.
Test on your own worst audio rather than a clean sample. Every tool performs well on clean audio.
Key takeaways
- Accurate on clear single-speaker audio, unreliable on crosstalk and phone bridges.
- Accuracy varies by accent, which is a fairness issue where transcripts affect people.
- Say it out loud rather than relying on a notification banner.
- Cloning a real voice needs consent in writing and usually a licence.
- Test tools on your own worst audio, not a clean sample.
Frequently asked questions
How accurate is AI transcription?
Very accurate on clear single-speaker audio. It degrades with crosstalk, phone audio, specialist vocabulary and some accents, usually without flagging the loss.
Does it work with regional accents?
Accuracy varies across UK accents. Where transcripts feed a record that affects someone, that unevenness needs a human check rather than trust.
Do I need consent to record?
You should tell people before recording, out loud rather than relying on a banner, and particularly with external participants.
Where is the audio stored?
Many tools process externally by default. Check the location and retention before adopting one, because that decides more than accuracy for most organisations.
Can I improve accuracy?
Add a custom vocabulary with your product names and colleagues' surnames. That removes the commonest error class immediately.
Is synthetic voice acceptable at work?
For internal narration and accessibility where no individual is implied, yes. For anything presented as a named person, only with written consent.
What is transcription best for?
Interviews and research calls, dictating first drafts, accessibility captions, and producing translatable text from audio.
How do I choose a tool?
Test on your own worst audio rather than a clean sample. Every tool performs well on clean audio, which tells you nothing.
The funded route, in figures
Figures are the England position. Levy-paying employers usually draw the whole cost from the levy they already contribute, so they pay nothing extra. Employers below the £3 million pay bill threshold pay nothing for an apprentice aged 16 to 24. For an older apprentice the government funds 95 per cent, and a levy transfer from a larger employer can cover the rest. Eligibility and funding rules are set by the funding body and can change.
Where to go next
If you want tool selection and responsible use assessed formally, the funded route covers both on your own work.
We help UK employers and employees access government-funded AI training. Training is delivered by third-party providers. We do not guarantee a place, a qualification, a job or any particular outcome. Eligibility and funding rules are set by the funding body and can change. Always check the current position before you commit. See our disclaimer.