app.py

import whisper
import moviepy.editor as mp
import gradio as gr
import torch
import subprocess

# 1. Load Whisper Model and Move to GPU if Available
device = "cuda" if torch.cuda.is_available() else "cpu"
model_name = "tiny"  # Use 'tiny' model for faster processing, change to 'base' or larger for more accuracy
whisper_model = whisper.load_model(model_name).to(device)

# Helper function to convert Whisper's transcription to SRT format
def generate_srt(transcription_result):
    """Convert transcription with timestamps to SRT format."""
    srt_content = ""
    for i, segment in enumerate(transcription_result['segments']):
        start = segment['start']
        end = segment['end']
        start_time = f"{int(start//3600):02}:{int((start%3600)//60):02}:{int(start%60):02},{int((start%1)*1000):03}"
        end_time = f"{int(end//3600):02}:{int((end%3600)//60):02}:{int(end%60):02},{int((end%1)*1000):03}"

        srt_content += f"{i+1}\n{start_time} --> {end_time}\n{segment['text'].strip()}\n\n"
    return srt_content

# 2. Function to Extract Audio using optimized FFmpeg subprocess
def extract_audio_ffmpeg(video_file, audio_output):
    """Extract audio using ffmpeg with optimized settings."""
    subprocess.run([
        'ffmpeg', '-i', video_file, '-vn', '-acodec', 'pcm_s16le', '-ar', '16000', audio_output, '-y'
    ])

# 3. Function to Transcribe, Translate (if necessary), and Generate Subtitles
def transcribe_and_generate_subtitles(video):
    # Step 1: Extract audio from video using optimized FFmpeg
    audio_path = "temp_audio.wav"
    extract_audio_ffmpeg(video, audio_path)

    # Step 2: Transcribe the audio
    transcription_result = whisper_model.transcribe(audio_path, language="en", verbose=False)
    
    # Step 3: Detect language and translate if necessary
    detected_language = transcription_result['language']
    if detected_language == "hau":  # Assuming "hau" is the code for Hausa
        # Re-transcribe with translation to English
        transcription_result = whisper_model.transcribe(audio_path, task="translate", verbose=False)

    # Step 4: Generate SRT subtitles
    srt_content = generate_srt(transcription_result)

    # Save SRT file
    srt_file = "output_subtitles.srt"
    with open(srt_file, "w", encoding="utf-8") as f:
        f.write(srt_content)

    # Step 5: Add subtitles to the video
    output_video = "video_with_subtitles.mp4"
    subprocess.run([
        'ffmpeg', '-i', video, '-vf', f"subtitles={srt_file}", output_video, '-y'
    ])

    return transcription_result["text"], output_video

# 4. Set up Gradio Interface with Optimized Settings
interface = gr.Interface(
    fn=transcribe_and_generate_subtitles,
    inputs=gr.Video(label="Upload Video File"),
    outputs=[
        gr.Textbox(label="Transcription or Translation"),
        gr.File(label="Download Video with Subtitles")
    ],
    title="Video Subtitle Generator",
    description="Upload a video in either English or Hausa. The system will detect the language, transcribe or translate if necessary, and generate a video with subtitles embedded.",
    live=False  # Disable live updates for faster processing
)

# Launch the interface
interface.launch()