app.py

import gradio as gr
import time
import torch
import scipy.io.wavfile
from espnet2.bin.tts_inference import Text2Speech
from espnet2.utils.types import str_or_none
from espnet2.bin.asr_inference import Speech2Text

# tagen = 'kan-bayashi/ljspeech_vits' 
# vocoder_tagen = "none" 

speech2text = Speech2Text.from_pretrained(
    asr_train_config="slurp/config.yaml",
    asr_model_file="slurp/valid.acc.ave_10best.pth",
    # Decoding parameters are not included in the model file
    nbest=1
)
# Confirm the sampling rate is equal to that of the training corpus.
# If not, you need to resample the audio data before inputting to speech2text
# speech, rate = soundfile.read("audio--1504190171-headset.flac")
# nbests = speech2text(speech)

# text, *_ = nbests[0]
# print(text)
# exit()

# text2speechen = Text2Speech.from_pretrained(
#     model_tag=str_or_none(tagen),
#     vocoder_tag=str_or_none(vocoder_tagen),
#     device="cpu",
#     # Only for Tacotron 2 & Transformer
#     threshold=0.5,
#     # Only for Tacotron 2
#     minlenratio=0.0,
#     maxlenratio=10.0,
#     use_att_constraint=False,
#     backward_window=1,
#     forward_window=3,
#     # Only for FastSpeech & FastSpeech2 & VITS
#     speed_control_alpha=1.0,
#     # Only for VITS
#     noise_scale=0.333,
#     noise_scale_dur=0.333,
# )


# tagjp = 'kan-bayashi/jsut_full_band_vits_prosody' 
# vocoder_tagjp = 'none'

# text2speechjp = Text2Speech.from_pretrained(
#     model_tag=str_or_none(tagjp),
#     vocoder_tag=str_or_none(vocoder_tagjp),
#     device="cpu",
#     # Only for Tacotron 2 & Transformer
#     threshold=0.5,
#     # Only for Tacotron 2
#     minlenratio=0.0,
#     maxlenratio=10.0,
#     use_att_constraint=False,
#     backward_window=1,
#     forward_window=3,
#     # Only for FastSpeech & FastSpeech2 & VITS
#     speed_control_alpha=1.0,
#     # Only for VITS
#     noise_scale=0.333,
#     noise_scale_dur=0.333,
# )

# tagch = 'kan-bayashi/csmsc_full_band_vits'
# vocoder_tagch = "none" 

# text2speechch = Text2Speech.from_pretrained(
#     model_tag=str_or_none(tagch),
#     vocoder_tag=str_or_none(vocoder_tagch),
#     device="cpu",
#     # Only for Tacotron 2 & Transformer
#     threshold=0.5,
#     # Only for Tacotron 2
#     minlenratio=0.0,
#     maxlenratio=10.0,
#     use_att_constraint=False,
#     backward_window=1,
#     forward_window=3,
#     # Only for FastSpeech & FastSpeech2 & VITS
#     speed_control_alpha=1.0,
#     # Only for VITS
#     noise_scale=0.333,
#     noise_scale_dur=0.333,
# )

def inference(wav,lang):
  with torch.no_grad():
      if lang == "english":
          speech, rate = soundfile.read("audio--1504190171-headset.flac")
          nbests = speech2text(speech)
          text, *_ = nbests[0]
      # if lang == "chinese":
      #     wav = text2speechch(text)["wav"]
      #     scipy.io.wavfile.write("out.wav",text2speechch.fs , wav.view(-1).cpu().numpy())
      # if lang == "japanese":
      #     wav = text2speechjp(text)["wav"]
      #     scipy.io.wavfile.write("out.wav",text2speechjp.fs , wav.view(-1).cpu().numpy())
  return  text
title = "ESPnet2-SLU"
description = "Gradio demo for ESPnet2-SLU: Extending the Edge of SLU Research. To use it, simply record your audio. Read more at the links below."
article = "<p style='text-align: center'><a href='https://github.com/espnet/espnet' target='_blank'>Github Repo</a></p>"

examples=[['audio-_slurp.flac',"english"]]

# gr.inputs.Textbox(label="input text",lines=10),gr.inputs.Radio(choices=["english"], type="value", default="english", label="language")
gr.Interface(
    inference, 
    [gr.inputs.Audio(label="input audio"),gr.inputs.Radio(choices=["english"], type="value", default="english", label="language")], 
    gr.outputs.Textbox(type="str", label="Output"),
    title=title,
    description=description,
    article=article,
    enable_queue=True,
    examples=examples
    ).launch(debug=True)