My Voice Was Cloned Before Lunch on Day One
Friday fun: my first week at a voice AI company has been going well, but nothing prepared me for the moment someone cloned my voice and played it back at me.
Voice cloning on day one
I started at Speechify on Monday. By Tuesday morning I had my laptop, my Slack access, and a note from an engineer saying “can you record a quick sample so we can test the voice cloning pipeline with a British accent?”
Sure. I found a quiet room. Recorded thirty seconds of a passage from a book I had on my phone. Sent the file. Forgot about it.
Wednesday afternoon I was in a meeting when someone said “hey, we loaded your clone. Want to hear it?” They played a sentence. It was my voice. It said something about streaming TTS latency benchmarks. I recognised the rhythm, the way I drop the ends of words, the specific hesitation I do before long nouns. It was me. It was also absolutely not me. The uncanny feeling sat somewhere between hearing your own voicemail greeting and watching a deepfake of yourself order a pizza.
The weirdest part came Thursday. I was writing a guide and needed to test an API call. I used a shared voice ID from the catalog. The output was fine. Then I swapped in my cloned voice ID and ran the same sentence again. My own voice read my own documentation back to me. In a guide I was writing for other developers. It felt like being in a room with an echo that knew what I was going to say next.
The engineer who asked for the sample thought this was hilarious. “Everyone does this on their first week,” she said. “You get used to it.”
I have been in developer relations for years. I have written about APIs I never touched and products I never ran. But I have never had a product talk back to me in my own voice before. That part of onboarding is not in any handbook.
I did not get used to it by Friday. But I stopped flinching when it played. Small wins.
FAQ
How does Speechify voice cloning work?
Upload a 10-30 second audio sample of a person speaking clearly, with their consent. The API returns a voice ID you can use in any speech endpoint. The clone captures accent, tone, and speaking style. See the docs at docs.speechify.ai.
How long does it take to clone a voice?
The API returns a voice ID within a few seconds of receiving your sample. The first synthesis takes a bit longer because the model processes the voice profile server-side. After that, every subsequent call with that voice ID runs at normal speed. The once-per-voice setup cost is negligible for most use cases.
Is voice cloning safe?
Speechify requires consent on every clone. You send the speaker’s full name and email as part of the creation request. The feature is designed for legitimate use cases like accessibility, content creation, and personalisation, not impersonation.
Did I keep using my cloned voice?
I did not. It was a test sample and it served its purpose. But now I know exactly what my own documentation sounds like in my own voice. That is a specific kind of developer experience I did not expect to have.