I second Majel Barrett, I’d be willing to do the training if I knew how.
I replied to someone upthread who had exactly that, as well as some tricks for getting it working in piper
Took me forever to get these custom models working and showing up in the Piper list:
First things first, I had to make sure the dataset field matched the exact filename of the .onnx file:
{
"dataset": "BT7274",
...
}
After that, I renamed everything to match for my benefit:
Default voices get downloaded and put into this directory (since I have the Docker container), and they will have the funky en_US-xxx-medium names.
Last, I updated the JSON in every single one to match the quality and add the language info:
{
"dataset": "BT7274",
"audio": {
"sample_rate": 22050,
"quality": "medium"
},
"espeak": {
"voice": "en-us"
},
"language": {
"code": "en_US",
"family": "en",
"region": "US",
"name_native": "English",
"name_english": "English",
"country_english": "United States"
},
...
}
The sample_rate needs to match, but all of these were 22050Hz.
What I need to do check the filesize. Anything ~60MB is a medium quality and ~110MB is high. That made it easy to put in the right info into each one.
They all work now and show up in the dropdown 
.
Just wanted to say thanks for the addition of the Subnautica PDA voice! One minor note... I did need to add a "dataset": "pad" entry in the .onnx.json file to get it working for my use case. Thanks again!