happy0708/minnan_test

1import gradio as gr2import torch3from transformers import AutoTokenizer, VitsModel4 5 6MODEL_NAME = "facebook/mms-tts-nan"7 8print("Loading model...")9 10tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)11 12model = VitsModel.from_pretrained(MODEL_NAME)13model.eval()14 15 16def tts(text):17 18    text = text.strip()19 20    if not text:21        return None22 23    inputs = tokenizer(24        text,25        return_tensors="pt"26    )27 28    with torch.inference_mode():29        output = model(**inputs)30 31    waveform = output.waveform.squeeze().numpy()32 33    return (34        model.config.sampling_rate,35        waveform36    )37 38# -----------------------------39# Gradio UI40# -----------------------------41demo = gr.Interface(42    fn=tts,43    inputs=gr.Textbox(44        label="輸入文字",45        placeholder="請輸入閩南語文字"46    ),47    outputs=gr.Audio(48        label="生成語音",49        type="numpy"50    ),51    title="Taiwanese TTS",52    description="facebook/mms-tts-nan"53)54 55demo.queue()56 57demo.launch()