happy0708/minnan_test

1import tempfile2 3import gradio as gr4import soundfile as sf5 6from transformers import (7    AutoTokenizer,8    VitsModel9)10 11import torch12 13MODEL_NAME = "facebook/mms-tts-nan"14 15tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)16model = VitsModel.from_pretrained(MODEL_NAME)17 18 19def tts(text):20 21    inputs = tokenizer(text, return_tensors="pt")22 23    with torch.no_grad():24        output = model(**inputs)25 26    waveform = output.waveform.squeeze().cpu().numpy()27 28    file = tempfile.NamedTemporaryFile(29        suffix=".wav",30        delete=False31    )32 33    sf.write(34        file.name,35        waveform,36        model.config.sampling_rate37    )38 39    return file.name40 41 42demo = gr.Interface(43    fn=tts,44    inputs=gr.Textbox(label="輸入文字"),45    outputs=gr.Audio(type="filepath"),46    title="Taiwanese TTS"47)48 49demo.launch()