app.py

import torch
import torchaudio
import gradio as gr

device="cpu"
bundle = torchaudio.pipelines.TACOTRON2_WAVERNN_PHONE_LJSPEECH
processor = bundle.get_text_processor()
tacotron2 = bundle.get_tacotron2().to(device)

# Workaround to load model mapped on GPU
# https://stackoverflow.com/a/61840832
waveglow = torch.hub.load(
    "NVIDIA/DeepLearningExamples:torchhub",
    "nvidia_waveglow",
    model_math="fp32",
    pretrained=False,
)
checkpoint = torch.hub.load_state_dict_from_url(
    "https://api.ngc.nvidia.com/v2/models/nvidia/waveglowpyt_fp32/versions/1/files/nvidia_waveglowpyt_fp32_20190306.pth",  # noqa: E501
    progress=False,
    map_location=device,
)
state_dict = {key.replace("module.", ""): value for key, value in checkpoint["state_dict"].items()}

waveglow.load_state_dict(state_dict)
waveglow = waveglow.remove_weightnorm(waveglow)
waveglow = waveglow.to(device)
waveglow.eval()

def inference(text):

  with torch.inference_mode():
      processed, lengths = processor(text)
      processed = processed.to(device)
      lengths = lengths.to(device)
      spec, _, _ = tacotron2.infer(processed, lengths)



  with torch.no_grad():
      waveforms = waveglow.infer(spec)

  torchaudio.save("_assets/output_waveglow.wav", waveforms[0:1].cpu(), sample_rate=22050)
  return "output_waveglow.wav"
  
gr.Interface(inference,"text",gr.outputs.Audio(type="file")).launch(debug=True)