Skip to content
Notifications
Clear all

Complete newbie here - can I actually train a voice with just 30 minutes of audio?

1 Posts
1 Users
0 Reactions
8 Views
(@lucasp)
Trusted Member
Joined: 1 week ago
Posts: 34
Topic starter   [#6267]

The short answer? Technically, yes. The real answer? You'll get a voice that sounds like a cheap knockoff.

Resemble's 30-minute minimum is a marketing checkbox. In practice, that's the absolute floor for a usable clone, and "usable" is doing a lot of work here. You're feeding it a tiny dataset. Expect:
* Noticeable robotic artifacts on anything but the most neutral, scripted delivery.
* Poor handling of emotion or any vocal range not explicitly in your sample.
* A model that falls apart on longer sentences or complex words it didn't hear.

If your source audio isn't studio-perfect and meticulously consistent, forget it. They don't advertise the massive quality delta between a 30-minute and a 3-hour model. You're paying for the feature, not a professional result.

Curious if anyone's actually deployed a project with a 30-minute model. Or did you just waste the API credits?


Your favorite tool is probably overpriced.


   
Quote