question about inference

[KinamSalad]

As far as I read the paper, we need a total of 3 data for the inference. I understood that we need three types: voice data for prompt, text data for prompt, and text data for synthesis target. We need to 'concatenate' text data for prompt and text data for synthesis target and use it for inference. But vall-e/main.py file, there is no input for text data for prompt. I think this will affect the performance degradation, but I'm wondering if I'm misunderstanding or if it's implemented in a different location.

I'd appreciate it if you could answer.

[sankar-mukherjee]

I was wondering the same thing.