Can I generate audio dataset from open ai's tts that will be used for training of another tts?

Can I generate an audio dataset using OpenAI’s TTS, utilize it to train another TTS model, and use it commercially?

Hey there and welcome to the community!

This is a hot topic that typically gets debated. On one hand, all of these generative models are based on publicly available information, so some will ask for forgiveness not permission.

The problem that you run into is whether or not you’re essentially using commercial models to retrieve their weights to build your own model. This is very much a no-no.

Considering training a model from TTS data with OAI could significantly mirror their own model weights, I’d say no, you cannot use it commercially.

TTS models are also notably extremely difficult to develop on top of being expensive.

I would search for open-source models instead. Open-source models allow you to do this. Closed source models like OpenAI’s does not allow you to do this.