metadata
language:
- zh
- tw
pipeline_tag: text-to-speech
base_model: Jackellie/ellie-Bert-VITS2
Taiwan accent TTS model from JackEllie.
Usage
Using this checkpoint from Hugging Face Transformers:
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())