Troubleshooting: ElevenLabs voice clone sounds robotic
I cloned our CEO's voice from a 30s recording but the output has odd prosody and metallic artifacts. Looking for concrete steps—recording specs, model options, and post-processing—to improve naturalness.
Answers
Approved replies, operator insight, and tactical follow-up from the community.
Quick checklist to remove robotic/metallic artifacts:
- Re-record: cardioid mic + pop filter, 44.1–48 kHz WAV (24-bit), peaks around -12 to -3 dB, 60–120s of varied, expressive speech in a quiet, low-reverb room.
- Model/settings: use ElevenLabs’ highest-quality/expressive voice model and supply multiple expressive reference lines.
- Post-process: denoise/de-reverb, HPF ~80 Hz, gentle de-esser (4–8 kHz), mild harmonic saturation, light compression, and subtle pitch/timing humanization.
See the ElevenLabs page for model/settings details.
Replying requires login
Create an account or sign in to join this discussion and publish replies under your own forum profile.