Here are the best-practice inputs for the most accurate Instant Voice Cloning (IVC):
Provide 1–2 minutes total of clean speech; avoid exceeding ~3 minutes. Total runtime matters more than the number of clips. (elevenlabs.io)
Use one speaker only, recorded in a quiet space with no background noise or reverb. Prefer a few consistently high‑quality clips over many mixed‑quality ones. If needed, lightly denoise before upload. (help.elevenlabs.io)
Keep performance consistent across samples (tone, pace, emotion, accent, breathing). The model will mimic whatever it hears—including mouth clicks and artifacts—so avoid dynamic level swings or performance shifts. (elevenlabs.io)
Target recording level around −23 to −18 dB RMS with a true peak near −3 dB; avoid clipping and overly quiet takes. (elevenlabs.io)
Format: MP3 works well. IVC accepts 128 kbps and above; Help Center recommends 192 kbps+. Higher bitrates yield little improvement compared to capture quality—prioritize clean recording. (elevenlabs.io)
If you need even higher fidelity and long‑term consistency, consider Professional Voice Cloning with substantially more speech (ideally ~2–3 hours, minimum ~30 minutes) and the same cleanliness/consistency requirements. (elevenlabs.io)
jolly ermine
craggy oracle