AI-generated synthetic voices: the identity question

When people imagine losing their voice, they picture not being able to speak. The harder loss is quieter than that. It is hearing a flat, generic computer voice say the words you typed and not recognizing yourself in them. The accent is gone. The warmth is gone. The particular way you said your kid’s name is gone. The words come out, but they do not sound like you, and to the people who love you, that gap is its own kind of absence.

Voice banking exists to close that gap. For people facing voice loss, whether from ALS, motor neurone disease, throat cancer, or surgery, the technology can preserve something generic synthetic speech never could: a voice that is recognizably theirs. It is one of the most personal uses of AI in disability, and it raises a question that runs deeper than the engineering. If a machine carries your voice forward, in what sense is it still yours?

What voice banking and message banking actually do

Two related approaches preserve a person’s voice, and the difference between them matters when you are deciding what to do.

Voice banking is the comprehensive one. You record a large set of sentences, sometimes hundreds or thousands, while your voice is still clear. Software samples your vocal qualities across many contexts and builds a synthetic voice modelled on your own. Later, paired with communication software, it can speak any message you type in a voice built from yours, instead of an off-the-shelf computer voice.

Message banking is simpler and more intimate. Rather than building a full synthetic voice, you record specific meaningful phrases: the things you say often, the sounds that are uniquely yours, a laugh, a term of endearment, the way you greet someone. Those recordings play back in your actual voice, carrying the exact tone and emphasis no synthetic model fully reproduces. Many people do both. Message banking for the phrases that matter most, voice banking for everything else.

The field has real history behind it. The speech scientist Rupal Patel founded VocaliD, a spin-out from her research at Northeastern University that pioneered personalized synthetic voices for text-to-speech systems, blending a person’s own vocal sound with donor recordings so that someone who could produce very little speech could still have a voice that fit them rather than a stranger’s. That work helped establish what is now a growing set of tools, from clinical message-banking programs to AI systems that can reconstruct a voice from a short sample. VocaliD itself was acquired by the AI firm Veritone in 2022, which is a useful reminder of something this article keeps coming back to: the company holding your voice can change hands, and you want to know what happens to your recordings when it does.

The identity question

Here is where it gets genuinely complicated, and worth sitting with rather than rushing past. A synthetic voice built from your recordings is, in a real sense, you. It carries your accent, your pitch, your patterns. People who use these voices often describe the relief of finally sounding like themselves again, of their identity staying intact in a way a robotic voice would have erased.

And it is also not quite you. It is a model. It will say things in your voice that you never said and might never have said. As AI voice tools get better at rebuilding a voice from very little audio, the questions multiply. Who controls the model of your voice once it exists? Could it be used without your consent? What happens to it after you are gone, and who decides? These are not reasons to avoid voice banking. They are reasons to go in with your eyes open and your terms set.

There is no single right answer to how much this should weigh on you. Some people find deep meaning in preserving their voice and do not worry much about the philosophical edges. Others want firm control over how the voice can be used. Both are valid. The point is that it is your voice and your decision, and the dignity in the whole enterprise comes from you holding the controls, not from the technology.

What to ask before you record

Voice banking often happens under time pressure, especially with a progressive condition, and that pressure can push people to record first and read the fine print never. A few questions are worth answering before you start, so the voice you preserve stays under your control.

  • Who owns the recordings and the synthetic voice built from them? Read this carefully. Ownership should rest with you or your family, not become a company asset you cannot retrieve.
  • Can I get the raw recordings back in a standard format? If a service folds or gets sold, you do not want your own voice locked inside it.
  • How is consent handled, and what would stop the voice being used in ways I did not approve? Ask what controls exist against misuse.
  • What happens to my voice model after I die, and who decides? Make this an explicit decision, in writing, rather than a default buried in terms of service.
  • Will this voice work with the communication device or software I will actually use? A beautiful voice that will not load onto your speech-generating device is a recording, not a solution. Check compatibility first.
  • How much time and stamina does recording take, and can I do it in sessions? For a progressive condition, recording earlier, while your voice is clearest, generally produces a better result. That is a practical timing point worth raising with your speech-language pathologist.

Where to get help

Voice and message banking are best done with a speech-language pathologist, ideally one experienced with augmentative and alternative communication. They can guide the process, help you choose between message banking and full voice banking, and make sure what you create will work with your communication system. If you are living with ALS or motor neurone disease, established programs such as Team Gleason and the ALS Association run voice-preservation services and can point you to vetted tools rather than leaving you to sort through options alone. Starting that conversation early, before voice changes are advanced, gives you the most to work with.

The takeaway

A synthetic voice built from your own is one of the rare technologies that preserves something irreplaceable: not just the ability to communicate, but the sound of you doing it. That is worth a great deal. It is also a model of your voice that will outlast the recording session and raise real questions about ownership, consent, and what happens to it later. Go in deliberately. Record while your voice is clearest, work with a speech-language pathologist, and pin down who owns and controls the voice before you build it. The technology can keep you sounding like yourself. Making sure the voice stays yours, on your terms, is the part that is up to you.

Sources

Similar Posts