app.py

import gradio as gr
import whisper
import torch

# Load the Whisper base model
device = "cuda" if torch.cuda.is_available() else "cpu"
model = whisper.load_model("base", device=device)

def transcribe(audio):
    # Load and process the audio file
    audio = whisper.load_audio(audio)
    audio = whisper.pad_or_trim(audio)
    mel = whisper.log_mel_spectrogram(audio).to(model.device)

    # Detect the language of the audio
    _, probs = model.detect_language(mel)
    language = max(probs, key=probs.get)
    print(f"Detected language: {language}")

    # Transcribe the audio
    options = whisper.DecodingOptions(fp16=torch.cuda.is_available())
    result = whisper.decode(model, mel, options)

    return result.text

# Create the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## Multilingual Speech-to-Text Transcription")
    with gr.Tab("Upload Audio"):
        audio_file = gr.Audio(source="upload", type="filepath", label="Upload your audio file")
        transcribe_button = gr.Button("Transcribe")
        transcription_output = gr.Textbox(label="Transcription")
    with gr.Tab("Record Audio"):
        audio_record = gr.Audio(source="microphone", type="filepath", label="Record your audio")
        record_button = gr.Button("Transcribe")
        record_output = gr.Textbox(label="Transcription")

    # Define button actions
    transcribe_button.click(transcribe, inputs=audio_file, outputs=transcription_output)
    record_button.click(transcribe, inputs=audio_record, outputs=record_output)

if __name__ == "__main__":
    demo.launch()