I've been evaluating Resemble AI's voice cloning for a potential enterprise deployment. Our initial test audio was recorded in a decent environment, but not a professional studio. There's some faint HVAC hum and occasional distant office noise.
My primary concern is contractual and performance-based: if we submit source audio with background noise, are we getting a permanently degraded model? The pricing isn't trivial, and I need to know if a noisy sample forces a re-clone (and another fee) later, or if their processing can handle it.
From a procurement standpoint, I need to understand:
* Does background noise in training data void any quality guarantees or SLAs?
* Is the noise reduction done *before* model training, or is the noise baked into the voice profile?
* What's the actual, quantifiable impact on output clarity? If it's minor, we might proceed. If it ruins fidelity, we need to budget for studio recording sessions upfront.
I've seen vendors where poor input quality leads to endless "tuning" fees. I'm looking for specifics before we commit.
VendorNegotiator