Should We Really “Fix” That? AI, Authenticity and the Awkward Bits of Human Speech

One of the things we love about our work is that we deal with real people. People who say “um.” People who mis-speak. People who don’t quite trot out the exact words from the corporate comms playbook. Most of the time, we sort this on location - a gentle nudge, a quick retake, the client nodding in approval. Job done.

Now, of course, AI gives us a new bag of tricks. We can smooth out sound, remove filler words, and even clone a voice so seamlessly you’d swear the contributor nailed it first time. Useful? Definitely. No more expensive re-shoots because someone said “Tuesday” when they meant “Thursday.” No more painstaking edit sessions cobbling together a missing word syllable by syllable. And when it works, it works brilliantly.

But here’s where it gets interesting. If we can correct every slip, should we? If we can clone a whole interviewee and make them sound like the voiceover of their dreams, is that still them? Or is it us playing ventriloquist?

Take speech differences, for example. If a contributor asks for their speech to be smoothed out, should we always do it? On one hand, it might make them easier to follow. On the other hand, we risk losing something of their individuality and authentic voice. Not so long ago, everyone on British television was expected to sound like the royal family. Surely we’ve moved beyond that idea of uniformity?

Layered on top of all this are the legal considerations - from data protection to emerging AI rules - which everyone, us and our clients included, is still working out in real time.

And then there are accents. Ah, accents. Here in the Westcountry, we know what comes with them. A Devon or Cornish twang can still (unfairly) be pegged as “country bumpkin.” Many contributors are shocked when they first hear themselves played back. We see ourselves in mirrors every day, but hearing your own voice is a whole different experience. Our MD knows this well: proudly Estuary, but in her early BBC days she ironed it flat with a dash of Queen’s English because back then, English regional accents were nowhere to be heard on air.

These days, it’s more accepted, even fashionable, so she’s gone back to Estuary, h-less and all. But what if a contributor says: “Make me sound less Janner”? Do we respect the request, or do we collude with the stereotype?

The truth is, we don’t have neat answers. AI is clever, but it’s also a bit weird. Sometimes it produces miracles, sometimes it produces… something else entirely. And for every practical fix (that errant date, the mangled job title), there’s a murkier decision about identity, authenticity, and trust.

So here’s where we stand. We’ll always talk to clients before using these tools. We’ll be cautious about where the line sits between tidying and tampering. 

This isn’t a grand conclusion. It’s a reflection - and an invitation. What would you want if it was your voice on screen? To sound smoother? Or to sound like you?