app.py

import os
import gradio as gr

SRC_LIST = ['cs', 'de', 'en', 'es', 'et', 'fi', 'fr', 'hr', 'hu', 'it', 'nl', 'pl', 'pt', 'ro', 'sk', 'sl']
TGT_LIST = ['en', 'fr', 'es']
MODEL_LIST = ['xm_transformer_sm_all-en']
for src in SRC_LIST:
    for tgt in TGT_LIST:
        if src != tgt:
            MODEL_LIST.append(f"textless_sm_{src}_{tgt}")

examples = []

io_dict = {model: gr.load(f"huggingface/facebook/{model}") for model in MODEL_LIST}

def inference(audio, model):
    out_audio = io_dict[model](audio)
    return out_audio

gr.Interface(
    inference,
    [gr.Audio(source="microphone", type="filepath", label="Input"),gr.Dropdown(choices=MODEL_LIST, type="value", label="Model")
],
    gr.Audio(label="Output")).queue().launch()