app.py

import os



os.environ["CURL_CA_BUNDLE"]=""

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
from parallel_wavegan.utils import download_pretrained_model


gos_text2speech = Text2Speech.from_pretrained(
  model_tag="https://huggingface.co/ahnafsamin/FastSpeech2-gronings/resolve/main/tts_train_fastspeech2_raw_char_tacotron_train.loss.ave.zip",
  vocoder_tag="parallel_wavegan/ljspeech_parallel_wavegan.v3"
)

hoogelandsters_text2speech = Text2Speech.from_pretrained(
  model_tag="https://huggingface.co/ahnafsamin/FastSpeech2-gronings-hoogelandsters/resolve/main/tts_train_fastspeech2_raw_char_tacotron_train.loss.ave.zip",
  vocoder_tag="parallel_wavegan/ljspeech_parallel_wavegan.v3"
)

westerkwartiers_text2speech = Text2Speech.from_pretrained(
  model_tag="https://huggingface.co/ahnafsamin/FastSpeech2-gronings-westerkwartiers/resolve/main/tts_train_fastspeech2_raw_char_tacotron_train.loss.ave.zip",
  vocoder_tag="parallel_wavegan/ljspeech_parallel_wavegan.v3"
)

oldambster_text2speech = Text2Speech.from_pretrained(
  model_tag="https://huggingface.co/ahnafsamin/FastSpeech2-gronings-oldambster/resolve/main/tts_train_fastspeech2_raw_char_tacotron_train.loss.ave.zip",
  vocoder_tag="parallel_wavegan/ljspeech_parallel_wavegan.v3"
)

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())
      elif lang == "gronings hoogelandsters":
          wav = hoogelandsters_text2speech(text)["wav"]
          scipy.io.wavfile.write("out.wav", hoogelandsters_text2speech.fs , wav.view(-1).cpu().numpy())
      elif lang == "gronings westerkwartiers":
          wav = westerkwartiers_text2speech(text)["wav"]
          wav = wav * 15
          scipy.io.wavfile.write("out.wav", westerkwartiers_text2speech.fs , wav.view(-1).cpu().numpy())
          #data, sr = librosa.load("output.wav")
          #factor = 2.0
          #data *= factor
          #sf.write("out.wav", data, sr)
          
      elif lang == "gronings oldambster":
          wav = oldambster_text2speech(text)["wav"]
          scipy.io.wavfile.write("out.wav", oldambster_text2speech.fs , wav.view(-1).cpu().numpy())

  return  "out.wav", "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", "gronings hoogelandsters", "gronings westerkwartiers", "gronings oldambster"], type="value", default="gronings", label="language")], 
    [gr.outputs.Audio(type="file", label="Output"), gr.outputs.File()],
    title=title,
    examples=examples
    ).launch(enable_queue=True)