Piper adds guides for people choosing a local speech voice for commercial work. Michael Hansen explains how to read each voice’s model card, follow the earlier models it was trained from, and check the licenses through that chain. A separate list identifies voices whose full training chain uses recordings licensed for commercial use.
The guides distinguish replacement voices trained from recordings licensed for commercial use from older downloads based on restricted lessac material. Michael Hansen says the replacement voices are already published; older copies keep their original training history, so the relevant model card is the one supplied with the download. The guides leave open whether a trained model is a derivative work.