What is the fastest way to clone your voice for an AI avatar with minimal recording?

The fastest documented voice-clone setup for an AI avatar is Google Omni's avatar calibration: a short face scan plus reading double-digit numbers aloud to the camera — no sentences, no paragraphs. From that minimal session the model infers intonation, pauses, and syllable handling, and in testing it produced a stronger voice replica than face replica.

Run the setup in two steps: complete the face calibration in Google Omni's avatar feature, then read a series of double-digit numbers to the camera. That number-reading session is the entire voice sample — the setup never asks for a scripted sentence or paragraph, which is what makes it the most minimal structured recording process currently documented for an avatar-integrated pipeline. Google Omni is also the only AI video model right now offering native avatar generation, so the voice and face capture happen in one pass rather than in separate tools.

Number reading works because the model extracts the components of your voice rather than your vocabulary: intonation, pause patterns, and how you handle syllables are all present in spoken numbers, and Omni infers the rest. Across 30+ test outputs evaluating the model, the voice side outperformed the face side — the calibration produced a decent facial replica but a noticeably stronger vocal one, so if you only have a minute to record, the voice is the part you can trust.

Record the calibration the way you would any voice sample: a quiet room, consistent distance from the microphone and camera, and your natural speaking pace and expressiveness. The model is sampling your cadence, so reading the numbers flatly or rushing them gives it less to infer from.

Understand the trade-off you are making. Zero-shot voice cloning workflows generally accept anywhere from a few seconds to about a minute of natural speech, and community guidance consistently holds that longer, more expressive samples raise fidelity. Number reading sits at the frictionless end of that spectrum — it trades phonetic richness for speed. For a talking-head avatar it holds up well; if you need highly expressive delivery, expect a minimal-input clone of any kind to be the faster-but-flatter option.

Once the clone exists, script around its delivery limits: keep each speaking clip to 6–7 seconds, which testing found to be the ceiling where lip sync performs consistently for a single speaker in frame, and keep one speaker per frame — multiple people talking in the same shot remains a weakness across every AI video model tested, not just Omni.

Watch some of these to see what works for you:

It was just double digit numbers that I was saying to the camera — it didn't ask me to say a sentence, read out a paragraph.

— invideo's creative team, on Google Omni's avatar voice calibration