Comment by cellis
Could this be used to train a text -> audio model? I'm thinking of an architecture that uses RVQ. Would RVQ still be necessary?
Could this be used to train a text -> audio model? I'm thinking of an architecture that uses RVQ. Would RVQ still be necessary?
I believe DDN is capable of handling TTS (text-to-speech) tasks, because with the text condition, the generation space is significantly reduced.
And it's recommended to combine it with an autoregressive model (GPT) for more powerful modeling capabilities.