app.py

import gradio as gr
import yt_dlp
import os
import shutil
from faster_whisper import WhisperModel

# --- 1. Model Setup (Turbo Settings) ---
model = None

def load_model():
    global model
    if model is None:
        print("📥 Loading Whisper Model (Base + Turbo Settings)...")
        # 'base' model with int8 quantization for speed/accuracy balance
        model = WhisperModel("base", device="cpu", compute_type="int8")
        print("✅ Model Loaded!")
    return model

# --- 2. Logic: Download Audio from URL ---
def download_audio_from_url(url):
    output_path = "downloaded_audio"
    # Cleanup old files
    if os.path.exists(f"{output_path}.mp3"): os.remove(f"{output_path}.mp3")

    ffmpeg_dir = "/usr/bin" # System Path for Hugging Face

    ydl_opts = {
        'format': 'bestaudio/best',
        'outtmpl': output_path,
        'ffmpeg_location': ffmpeg_dir,
        'postprocessors': [{
            'key': 'FFmpegExtractAudio',
            'preferredcodec': 'mp3',
            'preferredquality': '192',
        }],
        'quiet': True,
        'no_warnings': True,
        'nocheckcertificate': True,
        'http_headers': {
            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
            'Referer': 'https://www.tiktok.com/'
        }
    }

    try:
        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
            ydl.download([url])
        return f"{output_path}.mp3"
    except Exception as e:
        raise Exception(f"URL Download Error: {str(e)}")

# --- 3. Main Transcribe Function (Handles Both) ---
def transcribe_media(url_input, file_input):
    
    # Decide source: Priority given to File if both exist, else URL
    audio_file_path = None
    
    try:
        # CASE 1: File Upload
        if file_input is not None:
            print(f"📂 Processing Uploaded File: {file_input}")
            audio_file_path = file_input
            
        # CASE 2: URL Input
        elif url_input and url_input.strip() != "":
            print(f"🔗 Processing URL: {url_input}")
            audio_file_path = download_audio_from_url(url_input)
            
        else:
            return "⚠️ Error: Please provide either a Link or Upload a File."

        # --- Transcribe ---
        if not os.path.exists(audio_file_path):
             return "❌ Error: File not found."

        current_model = load_model()
        
        # Turbo Settings: beam_size=1 (Fast), vad_filter=True (Skip Silence)
        segments, _ = current_model.transcribe(
            audio_file_path, 
            beam_size=1, 
            vad_filter=True
        )
        
        text = " ".join([s.text for s in segments])
        return text.strip()

    except Exception as e:
        return f"❌ Error: {str(e)}"

# --- 4. Turbo UI with Tabs ---
css = """
.container {max-width: 900px; margin: auto;}
.gr-button-primary {background: linear-gradient(90deg, #1CB5E0 0%, #000851 100%); border: none; color: white;}
.tab-nav {font-weight: bold; font-size: 1.1em;}
"""

with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
    
    with gr.Column(elem_classes="container"):
        gr.Markdown("# 🚀 Turbo Transcriber (Link & Upload)")
        gr.Markdown("Paste a TikTok link **OR** upload an Audio/Video file to get the text.")

        with gr.Tabs():
            
            # TAB 1: Link
            with gr.TabItem("🔗 Paste Link"):
                url_in = gr.Textbox(label="TikTok / YouTube URL", placeholder="https://...")
                btn_url = gr.Button("🚀 Transcribe Link", variant="primary")

            # TAB 2: File Upload
            with gr.TabItem("📂 Upload File"):
                file_in = gr.Audio(label="Upload Audio or Video", type="filepath", sources=["upload", "microphone"])
                btn_file = gr.Button("📂 Transcribe File", variant="primary")

        # Output Area (Common for both)
        transcript_out = gr.Code(label="Transcript Result", language="markdown", interactive=False, lines=15)

    # --- Button Actions ---
    # Logic: Pass 'None' to the unused input
    btn_url.click(
        fn=transcribe_media, 
        inputs=[url_in, gr.State(None)], # Link diya, File None
        outputs=transcript_out
    )
    
    btn_file.click(
        fn=transcribe_media, 
        inputs=[gr.State(None), file_in], # Link None, File diya
        outputs=transcript_out
    )

demo.launch()