app.py

import gradio as gr
import time
import urllib.request
from pathlib import Path
import os
import torch
import scipy.io.wavfile
from espnet2.bin.tts_inference import Text2Speech
from espnet2.utils.types import str_or_none

gos_text2speech = Text2Speech.from_pretrained(
   model_tag="https://huggingface.co/wietsedv/tacotron2-gronings/resolve/main/tts_ljspeech_finetune_tacotron2.v5_train.loss.ave.zip",
   vocoder_tag="parallel_wavegan/ljspeech_parallel_wavegan.v3",
   device="cpu",
   threshold=0.5,
   minlenratio=0.0,
   maxlenratio=10.0,
   use_att_constraint=True,
   backward_window=1,
   forward_window=4,
)

def inference(text,lang):
  with torch.no_grad():
      if lang == "gronings":
          wav = gos_text2speech(text)["wav"]
          scipy.io.wavfile.write("out.wav", gos_text2speech.fs , wav.view(-1).cpu().numpy())

  return  "out.wav"

title = "GroTTS"
examples = [
  ['Ze gingen mit klas noar waddendiek, over en deur bragel lopen.', 'gronings']
]

gr.Interface(
    inference,
    [gr.inputs.Textbox(label="input text", lines=3), gr.inputs.Radio(choices=["gronings"], type="value", default="gronings", label="language")], 
    [gr.outputs.Audio(type="file", label="Output"), gr.outputs.File()],
    title=title,
    examples=examples
    ).launch(enable_queue=True)