app.py

import librosa
import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Tokenizer
import gradio as gr
from transformers import pipeline
import IPython.display as display
import soundfile as sf
def speech_text(audio_file):
  tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-base-960h")
  model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
  speech, rate = librosa.load(audio_file,sr=16000)
  display.Audio(audio_file, autoplay=True)
  print(rate)
  input_values = tokenizer(speech, return_tensors ='pt').input_values
  #Store logits (non-normalized predictions)
  logits = model(input_values).logits
  #Store predicted id's
  predicted_ids = torch.argmax(logits, dim =-1)
  transcriptions = tokenizer.decode(predicted_ids[0])
  return transcriptions
iface = gr.Interface(speech_text,inputs="audio",outputs="text",title='Sakil Transcription',description="Transcription")
iface.launch(inline=False)