app.py

import gradio as gr
from faster_whisper import WhisperModel

# 1. LOAD THE OPTIMIZED MODEL
# "small" is a great balance. You can try "medium" if you have a GPU.
# "int8" makes it run fast on CPU.
model_size = "small"
model = WhisperModel(model_size, device="cpu", compute_type="int8")

def generate_subtitles(audio_file, task_type, language_setting):
    if audio_file is None:
        return "Error: No file uploaded."

    print(f"Processing... Task: {task_type}, Language: {language_setting}")

    # 2. Configure Language Code
    # faster-whisper expects language codes like "hi", "pa", "en"
    lang_map = {
        "Auto Detect": None,
        "Hindi": "hi",
        "Punjabi": "pa",
        "Urdu": "ur",
        "Bengali": "bn",
        "Tamil": "ta",
        "Telugu": "te",
        "Marathi": "mr",
        "Gujarati": "gu",
        "English": "en"
    }
    
    selected_lang = lang_map.get(language_setting)

    # 3. RUN FASTER-WHISPER
    # It returns segments (the text) and info (the detected language)
    segments, info = model.transcribe(
        audio_file, 
        task=task_type, 
        language=selected_lang,
        beam_size=5
    )

    # 4. Convert to SRT Format
    srt_content = ""
    for i, segment in enumerate(segments):
        start = format_timestamp(segment.start)
        end = format_timestamp(segment.end)
        text = segment.text.strip()
        srt_content += f"{i+1}\n{start} --> {end}\n{text}\n\n"

    return srt_content

# Helper to format seconds -> 00:00:00,000
def format_timestamp(seconds):
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    seconds = seconds % 60
    milliseconds = int((seconds - int(seconds)) * 1000)
    return f"{hours:02}:{minutes:02}:{int(seconds):02},{milliseconds:03}"

# 5. The User Interface
with gr.Blocks() as demo:
    with gr.Row():
        file_input = gr.Audio(sources=["upload"], type="filepath", label="Upload Audio/Video")
        
        with gr.Column():
            task_input = gr.Dropdown(
                choices=["transcribe", "translate"], 
                value="transcribe", 
                label="Action"
            )
            lang_input = gr.Dropdown(
                choices=["Auto Detect", "Hindi", "Punjabi", "Urdu", "Bengali", "Tamil", "English"], 
                value="Auto Detect", 
                label="Language"
            )
    
    output_text = gr.Textbox(label="Result (.SRT)", lines=15)
    submit_btn = gr.Button("Generate Subtitles (Fast)", variant="primary")

    submit_btn.click(
        fn=generate_subtitles, 
        inputs=[file_input, task_input, lang_input], 
        outputs=output_text, 
        api_name="predict"
    )

demo.launch()