The basics
What is AI voice cloning?
AI voice cloning builds a digital model of a person voice from a short recording of them speaking. Once the model exists, you type any text and get speech in that voice. The model captures timbre, pacing and accent patterns — it is not your original audio being replayed in pieces.
How does AI voice cloning work?
In four stages: you upload clean speech (30 seconds to a few minutes); the system extracts a speaker embedding, a numeric fingerprint of the voice; a text-to-speech model combines that fingerprint with the phonemes of your new text; a neural vocoder renders the result as an audio waveform. Cloning usually takes seconds to two minutes, and each generation after that takes a few seconds.
What is neural voice cloning?
Neural here means the voice model is a neural network trained on large speech corpora, rather than the older concatenative approach that stitched recorded fragments together. Neural cloning needs far less source audio — 30 seconds is usable — and it generalises to new text instead of only replaying recorded phrases.
Cost and credits
How much does it cost to clone a voice?
100 credits per voice, charged once. The voice is then reusable at no extra cost. Credits are $1 = 50, packs start at $9, with no monthly fee, no subscription and no expiry.
How is text-to-speech priced?
By UTF-8 bytes of input text: 4,200 credits per 1,000,000 bytes. English is about one byte per character; Chinese and Japanese are three bytes per character. A 1,000-word English script is roughly 6,000 bytes — about 25 credits.
What is a credit, and do credits expire?
One credit is $0.02 when you buy a pack ($9 buys 450 credits). Credits never expire. There is no monthly fee or subscription — you are billed for what you use and nothing else.
Recording and output
How much audio do I need for a good clone?
30 seconds of clean single-speaker speech is the practical minimum for a recognisable clone; 1 to 3 minutes markedly improves similarity. Avoid background music, reverb, other speakers and clipping. Mono, 44.1 kHz or higher, without heavy noise reduction.
How long does cloning take?
Training usually takes a few seconds to a couple of minutes depending on sample length and the current queue. The voice is usable for generation as soon as it appears in your library.
Which languages are supported?
Synthesis covers 83 languages, and the voice model is language-independent — record once in English, then generate in Japanese, German or Spanish. Accent in the output follows the training audio, so a native-sounding clone needs source audio in that accent.
Rights and reliability
Is voice cloning legal?
It depends on whose voice and what you do with it. Cloning your own voice is generally fine. Cloning someone else requires their informed consent — many jurisdictions treat a voice as a protected identifier, and some, such as the ELVIS Act in Tennessee, grant explicit publicity rights. Never use a clone to impersonate, defraud or misrepresent. Resona requires you to hold the rights or consent for every voice you clone. For a deeper breakdown of consent and disclosure across jurisdictions — including cloning an actor's voice for a dub — see our guide on <a href="/blog/voice-cloning-rights-consent">voice cloning rights and consent for dubbing</a>.
Can I use cloned voices commercially?
Yes, for voices you own or have written consent to use. Commercial use does not change the credit price. You remain responsible for the underlying rights, and for disclosing synthetic speech where the law or the platform requires it — several platforms now require an AI-voice label.
What happens if a task fails?
Reserved credits are refunded automatically. Your usage log keeps the attempt with a REFUND status so you can see exactly what happened. You are never charged for output you did not receive.
Something else? Write to support@resonaio.com — we reply within 24 hours on business days.