from transformers import AutoModel, AutoProcessor
from scipy.io.wavfile import write
import torch

model_name = "BricksDisplay/ellie-Bert-VITS2"

model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)

with torch.no_grad():
    inputs = processor("你好", language="zh", return_tensors="pt")
    result = model(**inputs)
    result = result["waveform"]
    write("output.wav", model.config.sampling_rate, result[0].numpy())