app.py

import gradio as gr
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, WhisperForConditionalGeneration, WhisperProcessor

# Load the model and processor
model_id = "openai/whisper-medium"
processor = WhisperProcessor.from_pretrained(model_id)

model = WhisperForConditionalGeneration.from_pretrained(model_id)
model.config.forced_decoder_ids = None

def transcribelocal(microphone, file_upload):
  # Check which input is not None
  if microphone is not None:
    audio = microphone
  else:
    audio = file_upload

  # Use the processor to transcribe the audio
  transcription = processor.transcribe(audio, 48)
    
  # Extract the confidence score and the duration from the transcription
  confidence = transcription.confidence
  duration = transcription.duration
    
  # Remove the special tokens from the transcription text
  text = transcription.text.replace("<|startoftranscript|>", "").replace("<|endoftranscript|>", "")
    
  # Return the text, confidence and duration as outputs
  return text, confidence, duration

# Create a Gradio interface with two modes: realtime and file upload
iface = gr.Interface(
  fn=transcribelocal,
  inputs=[
    gr.inputs.Audio(source="microphone", type="numpy", label="Realtime Mode"),
    gr.inputs.Audio(source="upload", type="numpy", label="File Upload Mode")
  ],
  outputs=[
    gr.outputs.Textbox(label="Transcription"),
    gr.outputs.Textbox(label="Confidence Score"),
    gr.outputs.Textbox(label="Duration (seconds)")
  ],
  title="Whisper Transcription App",
  description="A Gradio app that uses OpenAI's whisper model to transcribe audio"
)

# Launch the app
iface.launch()