Short example where to put own data?

[MSDNAndi]

VALL-E works with very short speech samples.
I am staring at the repo and do not find the key point.
Where to put your own voice voice sample.

If somebody has a "minimal example" to follow, I would appreciate it.

[maxgoff]

Following the instructions, I am successfully getting training done now (examples/libritts). Up to Epoch 4 at the moment. I see in that directory a subdirectory ./prompts which contains 2 sets of .txt /.wav files. I am wondering if I substitute those specific files with my own files if that will result in my own voice samples being used to train. Thoughts?

[lifeiteng]

you can prepare your own dataset by using lhoste more info in #87, prompts are used in Inference not Training.

[maxgoff]

If I use my own voice as prompts during inference, will the output reflect the characteristics of my voice in the prompt files?
To personalize the output voice, does it require a full training set with my voice such as are provided for training purposes, or will the vocal characteristics from the prompt during inference be used? I can see where prompting may be for providing new text for processing. But if that's the case, then I don't see why the .wav files would be needed. In any event, thank you for this great work you are doing!

[chenjiasheng]

If I use my own voice as prompts during inference, will the output reflect the characteristics of my voice in the prompt files?

Yes.

To personalize the output voice, does it require a full training set with my voice such as are provided for training purposes, or will the vocal characteristics from the prompt during inference be used?

Training on a huge voice dataset including your own voice, or fine-tune the pretrained model with your own voice, can achieve better naturalness and similarity for your own voice.

But if that's the case, then I don't see why the .wav files would be needed.

The prompt .wav as well as the transcrpit is needed for inference, no matter whether you have trained or fine-tuned the model with your own voice.