facebook/speech_matrix

1import os2import gradio as gr3 4SRC_LIST = ['cs', 'de', 'en', 'es', 'et', 'fi', 'fr', 'hr', 'hu', 'it', 'nl', 'pl', 'pt', 'ro', 'sk', 'sl']5TGT_LIST = ['en', 'fr', 'es']6MODEL_LIST = ['xm_transformer_sm_all-en']7for src in SRC_LIST:8    for tgt in TGT_LIST:9        if src != tgt:10            MODEL_LIST.append(f"textless_sm_{src}_{tgt}")11 12examples = []13 14io_dict = {model: gr.load(f"huggingface/facebook/{model}") for model in MODEL_LIST}15 16def inference(audio, model):17    out_audio = io_dict[model](audio)18    return out_audio19 20gr.Interface(21    inference,22    [gr.Audio(source="microphone", type="filepath", label="Input"),gr.Dropdown(choices=MODEL_LIST, type="value", label="Model")23],24    gr.Audio(label="Output")).queue().launch()