Home Assistant .pt to .onnx file conversion

[JoeHogan]

Hi, i got a zip file that was created vaa xVA-Synth which contains several .pt files. I wan't to use these files as my Home Assistant Assist voice. I've tried converting then by including them in a pytorch script (the pt file extension checked out) but there is no model definition for the voice model files... the ai voice works perfectly if i import it into xVA-Synth, but i need an .onnx file.... how can i do that without the original model?

[JoeHogan]

Necermind... I think I got my answer... it looks like I can probably use this model: https://github.com/DanRuta/xVA-Synth/blob/master/python/fastpitch1_1/model.py

Haven't tried it yet, but guessing if I import that model and all its dependencies, I can load the file into pytorch, then export it as onnx... we shall see.

[JoeHogan]

I was wrong... or at least partly wrong... importing the .pt file i have (defined as a fastpitch 1.1 model) using your codebase produces errors... here is my code which imports xVA-Synth code (note: this is my 3rd day of python... im a node.js dev, so i may just be an idiot):

def main():

    logger = logging.getLogger('serverLog')
    PROD = False
    device = torch.device('cpu')

    models_manager = ModelsManager(logger, PROD, device)

    modelWrapper = FastPitch1_1(logger, PROD, device, models_manager)

    model = modelWrapper.model

    ckpt_path = './voice/character_voice.pt'
    ckpt = torch.load(ckpt_path, map_location="cpu")
    model.load_state_dict(ckpt)
    
    model.eval()

    dummy_input = torch.randn(1, 3, 224, 224)

    torch.onnx.export(model, dummy_input, './out/character_voice.onnx')

and here is the error i get when i hit the last line of the code above:

infer_advanced() got an unexpected keyword argument 'dur_tgt'

admittedly, i have no idea how to form the 'dummy_data' variable above... any guidance there would be great! That said, i don't think that's necessarily causing the error I'm getting...