var7/fastspeech

1from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub2from fairseq.models.text_to_speech.hub_interface import TTSHubInterface3import numpy as np4import gradio as gr5 6 7models, cfg, task = load_model_ensemble_and_task_from_hf_hub(8    "facebook/fastspeech2-en-ljspeech",9    arg_overrides={"vocoder": "hifigan", "fp16": False}10)11model = models[0]12#model = models13TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)14generator = task.build_generator(model, cfg)15 16 17def vocal(text):18    sample = TTSHubInterface.get_model_input(task, text)19    wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)20    return rate,wav21    22 23 24gr.Interface(25    fn=vocal, 26    inputs="text",27    outputs="audio"28).launch()29 30 31