app.py

import gradio as gr
from transformers import pipeline
from faster_whisper import WhisperModel
import librosa

# Load the model
# pipe = pipeline("automatic-speech-recognition", model="navidved/persian-whisper-large-v3-ct2")
model = WhisperModel("navidved/persian-whisper-large-v3-ct2", device="cpu", compute_type="int8")

# Define the maximum audio length in seconds
MAX_AUDIO_LENGTH = 40  # 40 seconds

# Define the inference function
def transcribe_audio(audio):
    if audio is None:
        return "No audio file uploaded. Please try again."
    
    try:
        audio_data, sr = librosa.load(audio, sr=None)
        duration = librosa.get_duration(y=audio_data, sr=sr)

        # Check if the audio is longer than the allowed duration
        if duration > MAX_AUDIO_LENGTH:
            return f"Audio is too long. Please upload an audio file shorter than {MAX_AUDIO_LENGTH} seconds."
        
        # Perform transcription
        segments, _ = model.transcribe(audio, vad_filter=True)
        result = segments.text
        return result
    except Exception as e:
        return f"Error during transcription: {str(e)}"

# Create a Gradio interface for uploading audio or using the microphone
with gr.Blocks() as interface:
    gr.Markdown("# Whisper Large V3 Speech Recognition")
    gr.Markdown("Upload an audio file or use your microphone to transcribe speech to text.")
    
    # Create the input and output components
    audio_input = gr.Audio(type="filepath", label="Input Audio")
    output_text = gr.Textbox(label="Transcription")
    
    # Add a button to trigger the transcription
    transcribe_button = gr.Button("Transcribe")
    
    # Bind the transcribe_audio function to the button click
    transcribe_button.click(fn=transcribe_audio, inputs=audio_input, outputs=output_text)

# Launch the Gradio app
interface.launch()