Most brands find out their voice clone sounds off after they've already sent the client the first draft. The mistakes that ruin a clone almost always happen before a single line of AI training ever runs — in the room, on the day, with the microphone already set up wrong.
We've built voice clones for brand narration, ad voiceovers, and podcast segments, and the pattern is consistent: the clone is only as good as the source audio. This is the checklist we actually run through before anyone presses record.
Why Recording Day Decides Everything
Voice cloning tools don't fix bad audio, they learn from it. ElevenLabs states this directly: the quality ceiling is set by the reference audio. Background noise or compression artifacts will influence the output.
There are two cloning paths:
- Instant Voice Cloning (IVC) works from short samples. Approximately 1-2 minutes of clear audio without any reverb or background noise is recommended. Avoid recording more than 3 minutes, as this can be detrimental to the clone. The AI mimics everything, including speed, inflections, breathing patterns, and mouth clicks.
- Professional Voice Cloning (PVC) is the one you want for a brand using this voice across years of content. The recommended amount of audio is 2-3 hours (30 minutes minimum). PVC requires a single, uninterrupted speaker throughout. For a detailed layout of the two tiers, check out our comparison of instant vs professional voice cloning.
| Metric / Tier | Instant Voice Cloning | Professional Voice Cloning |
|---|---|---|
| Audio needed | 1-3 minutes | 30 mins minimum, 2-3 hours ideal |
| Turnaround | Immediate, no training | Minutes to process, fine-tuning |
| Best for | Prototypes, quick tests | Long-term brand voice |
| Consistency | Lower | Higher |
| Plan requirement | Most plans | Creator plan or above |
Phase 1: Before the Mic Goes On
- Book the room, not just the talent. Make sure there's only a single speaking voice throughout the audio.
- Treat the space, don't just soundproof it. Treatment stops the room's own reflections from smearing the voice. Speaking close to a cardioid mic helps overpower the room.
- Set recording specs before anyone speaks. The consistent standard is 24-bit / 48kHz WAV or AIFF. ElevenLabs recommends WAV files at 44.1kHz or 48kHz with a bitrate of at least 24 bits. For details on the signal-to-noise thresholds needed, check out our guide on why 30 minutes of clean audio beats 3 hours of noisy recordings.
- Get the consent paperwork signed first. Consent should be documented in writing before recording starts.
Phase 2: During the Session
- Warm up before the take. A short run-through of the actual script settles pacing.
- Keep mic distance fixed. The standard range is 8-12 inches in front of the talent's mouth. Moving back and forth produces an inconsistent clone.
- Keep performance style consistent. If the voice needs to handle both calm explanation and high-energy ads, the script should include samples of both.
- Stay hydrated. Dry mouth produces clicks and smacks that show up in generated output.
Phase 3: After You Stop Recording
- Don't over-process before uploading. Heavy compression or aggressive EQ before training can strip out the vocal texture the model needs to learn.
- Check total runtime. Split multi-hour files into shorter chunks rather than uploading one massive file.
- Complete voice verification using the same setup you recorded with to avoid verification failure.
- To see how these files behave once uploaded, read our guide on what the stability, similarity, and style settings actually control.
The Disclosure Step Most Brands Skip
Using voice clones in market-facing content in 2026 carries disclosure obligations:
- New York's synthetic performer law requires conspicuous disclosure when an AI-generated figure that appears human shows up in a commercial advertisement.
- FTC guidance draws a clear line: substantive changes — face swaps, voice clones, AI lip-sync — require disclosure.
- Cint research found that 63% of U.S. consumers believe brands have a duty to disclose AI use in marketing.
A plain, visible label like "AI-generated voice" solves the requirement cleanly.
Our Session Checklist
- Confirm the cloning path (instant vs. professional).
- Lock room and gear (treated space, 24-bit/48kHz WAV, consistent gain).
- Sign consent in writing before recording.
- Script for range (calm narration and higher energy).
- Verify with matching equipment.
- Plan the disclosure label before the content ships.
A brand voice clone rarely exists in isolation. To see how this fits into an organic video channel, see our guide on building a high-performance YouTube content system.
Summary
A voice clone isn't a quick upload. It's a recording session with technical specs, a consent process, and a disclosure obligation.
Ready to build a brand voice clone that actually holds up? Get a free consultation and we'll walk through what your specific use case needs before you record.
FAQs
How long does a voice recording need to be for AI cloning?
Instant cloning needs 1-3 minutes. Professional Voice Cloning needs a minimum of 30 minutes, with 2-3 hours recommended.
Can I use my phone to record audio for a voice clone?
Yes, for Instant Voice Cloning. For Professional Voice Cloning, a proper microphone and interface recording at 24-bit/48kHz WAV is recommended.
Do I need consent to clone someone else's voice?
Yes, always. Platforms require verification, and state laws carry penalties for unauthorized voice cloning.
Do I have to disclose that a voice in my ad is AI-generated?
Yes. FTC guidance and state-level laws (such as New York's) require conspicuous disclosure for synthetic voices.
What's the difference between Instant and Professional Voice Cloning?
Instant cloning is fast but less consistent. Professional cloning requires hours of audio and fine-tuning but delivers higher fidelity.
Why does my AI voice clone sound inconsistent between videos?
This usually traces back to variable mic distance, mixed recording sessions, or reference audio that only captured one tone.
Can background noise ruin a voice clone?
Yes. Models learn everything in the audio, including room echo, laptop fans, and mouth clicks.
Is a home setup good enough for a professional voice clone?
Yes, if the room is treated to prevent echo, mic distance is fixed, and recording specs meet the WAV standard. Treatment matters more than gear.