The Quick Answer
We do not recommend you clone your voice at this stage, if you're providing more than 3 or 4 minutes of audio to do so.
Cloning based on more than 3 or 4 minutes is probably considered a 'professional clone of your voice'.
We think this advice is really important if you have a regional accent. You could be basically giving the Text to Speech system what it needs to compete with yourself in the future by cloning your accent - even if it doesn't use your exact likeness.
If you are intrigued as to how your clone might sound, then we'd be happy to do it on your behalf.
More In Depth Answer
Companies are creating AI voice models using lots of training data as you know.
When a quick-clone of yourself is generated, the system reads your 2-3 minute audio file and converts it to a location in multi-dimensional space. If there's already data in this part of their model then it will use this data and your quick-clone will sound quite accurate. If there isn't data in this location it will essentially have to guestimate a bit more, and it won't be such a reliable clone. You may even end up sounding American if you're British and vice versa.
To alleviate the risks of a quick-clone being bad, many providers can do a 'professional clone' of your voice. They will take the normal model and add in your training data - but this time there will be much more than 2-3 minutes. They will be using your voice data to populate that 'area' of the training model that is missing. Your timbre is clustered around a similar location in multi-dimensional space.
Then they can do the same procedure as a quick-clone and it will come out sounding much more like you.
So to reiterate, when you create a professional clone you provide a lot more audio data, and your data is basically adding to the capabilities of the model. Do you trust these companies to keep this data for just your use?
So do you clone your voice? If you want to test how your clone would sound then we'd be happy to assist in this process. If you do it yourself then our recommendations is to avoid giving over too much of your voice data.
What About Cloned Voice Marketplaces
We’re also seeing more websites acting as AI voice marketplaces. These platforms are often marketed as a way for voice artists to license and monetise their voice, some even allow you to check which scripts you allow your cloned voice to be used on, and set your own rates. However the real concern is who is providing the underlying AI technology and whether the companies cloning your likeness can truly be trusted.
Once your voice is cloned and made available in a marketplace, how much control does the original artist actually retain? Is there a guarantee that your cloned voice won’t later be absorbed into future AI model training? Some AI providers are known to continuously refine and expand their datasets - meaning a voice artist’s likeness could ultimately end up being used in ways they never agreed to.
Also by offering your voices on these platforms are you legitimising the use of AI voices. A clone often only clones your likeness and not your other abilities which make you a great voiceover artist - if these platforms get popular you will start to get competition from people with a good vocal tone but without all the other skills of being a voiceover artist - but this won't matter. Then there will be a downward spiral in rates.
There's something clients should also be aware of when using these services. Many businesses are now uploading scripts directly into AI voice platforms, often without considering where that data goes. AI companies place immense value on data collection, and there’s a risk that sensitive, confidential, or proprietary content could be retained and used to improve future AI models. Clients should think carefully before trusting AI-generated voice services with their scripts.
