Voice AI: transcription, dubbing and the practical wins for small businesses
Voice AI has gone from sci-fi demo to genuinely useful in 18 months. Three branches of it - transcription, voice cloning, dubbing - all have practical applications for SMEs.
Most AI conversation focuses on text - ChatGPT, Claude, Gemini handling the writing work. The voice side has caught up quickly in the last 18 months. For some businesses, the voice tools save more time than the text ones.
This article covers the three branches of voice AI worth knowing about, where each one is practically useful for SMEs in North Wales, and the ethical points worth thinking about.
The three branches of voice AI
Transcription turns audio into text. Voice notes, recorded meetings, podcasts, customer calls. The leader is Whisper (made by OpenAI, free open-source model, also embedded in many products). Accuracy is excellent on clear audio in major languages including Welsh.
Voice cloning and synthesis turns text into spoken audio. The leader is ElevenLabs. You can train a voice on a few minutes of recorded sample, then have it read any text. Quality is now indistinguishable from real recording in many cases.
Dubbing combines transcription, translation and voice cloning to convert spoken audio from one language to another while preserving the speaker's voice and lip-sync. Newer than the other two, getting better fast.
Where transcription works for SMEs
Transcription is the most immediately useful branch for most North Wales businesses.
Meeting notes. Record a meeting (with consent), get an automatic transcript and summary. Most major meeting tools (Teams, Google Meet, Zoom, Otter.ai) now have this built in. The owner does not have to take notes; the team gets a searchable record. Built on RAG for follow-up "what did we agree about X?" questions.
Voice notes. Record yourself thinking out loud on a walk, get a transcript, paste it into ChatGPT to turn into a polished email or document. This is the single biggest workflow change for many business owners I work with - it shifts the writing from "sat at desk" to "while doing other things".
Customer interviews. Recorded sales calls, support calls or research interviews become searchable. You can ask AI to find patterns: "what is the most common objection?" "what features did the last 10 customers ask about?". Major sales tools (Gong, Chorus, HubSpot Conversations) do this natively.
Welsh transcription. Whisper handles Welsh well in 2026 - good enough for transcription work. The Welsh and AI post covers the broader picture.
Where voice cloning fits
More niche than transcription but powerful for specific use cases.
Audio versions of written content. A blog post, a newsletter, a guide - if your audience listens more than reads, AI voice can produce an audio version in minutes that sounds like a real narrator. Useful for accessibility and for content reuse.
Personalised voice in customer comms. Some businesses use a cloned founder voice for welcome messages, automated reminders or onboarding videos. The intent is warmth at scale - the customer hears your actual voice in a personalised message even though the message was generated.
Audiobooks and training. If you produce training material, the cost of professional narration was previously prohibitive for small audiences. AI voice has changed the maths - £20-50 of AI time produces what used to cost £500-1000 of studio recording.
Where dubbing matters
If your business produces video content (training, marketing, product demonstrations) and serves multilingual audiences, dubbing is genuinely transformative.
A 5-minute training video that needed three weeks of voice talent to produce in three languages can now be dubbed by AI in under an hour. The output is not yet perfect for marketing-grade content, but for internal training and informational content it is more than good enough.
For a North Wales business serving Welsh and English audiences, this is an obvious win. For businesses with EU or international customers, it expands what is feasible.
The ethical angle
Voice AI raises ethical questions that text AI does not.
Consent. You should not clone someone's voice without their consent. Most voice-cloning services require this contractually. Beyond the legal issue, the social one is real - using someone's voice without permission is closer to identity theft than to writing in their style.
Disclosure. If you publish AI-generated voice content, label it. The EU AI Act requires this for deepfakes; common decency requires it for any case where the listener might be misled. Transparency is the lower-cost choice.
Recording consent. Transcription requires recording. Recording requires consent. UK law generally requires both parties to be aware they are being recorded. Build the disclosure into your meeting setup.
Cost
Cheaper than people expect.
Transcription: free for many SME use cases. Whisper is free open-source. Built-in features in Teams, Meet, Zoom are bundled into existing licensing. Otter.ai and similar are around £15-30 a month.
Voice cloning: ElevenLabs starts at around £20 a month, scales with usage. For typical SME use (a few audio versions of written content per month), the lower tiers are sufficient.
Dubbing: more expensive at scale but per-video costs are typically £10-50 for a few minutes of dubbed content. A fraction of human dubbing.
The honest summary
For most North Wales SMEs, the voice branch worth starting with is transcription. The voice-note-to-document workflow alone changes how an owner can work. The meeting transcription saves real time. The cost is often zero on tools you already pay for.
Voice cloning and dubbing are more specialised. Worth knowing about; revisit when a specific use case appears.
If you would like to think through where voice AI might fit your business, that is what a discovery call is for. The 30-day plan post sets the broader adoption pattern.