subtitle-generator 3.0.0

AI-powered subtitle generation from video/audio using Whisper.

Subtitle Generator

AI-powered subtitle generation using Whisper for accurate speech-to-text transcription.

Features

• 🎯 Multi-format output - VTT, SRT, TXT, JSON, LRC, ASS, TTML
• 🚀 Fast processing - Powered by whisper.cpp for high-performance inference
• 📦 Batch processing - Process multiple videos at once
• 🔄 Video embedding - Embed subtitles directly into videos
• 🌍 Multilingual - Support for multiple languages

Installation

pip install subtitle-generator

Prerequisites

• FFmpeg is required for video/audio processing:

  # macOS
  brew install ffmpeg
  
  # Ubuntu/Debian
  sudo apt install ffmpeg
  
  # Windows (via chocolatey)
  choco install ffmpeg
  

Quick Start

# Generate subtitles (VTT format)
subtitle video.mp4

# Generate SRT format
subtitle video.mp4 --format srt

# Embed subtitles into video
subtitle video.mp4 --merge

# Use a larger model for better accuracy
subtitle video.mp4 --model large

CLI Commands

Command	Description
subtitle <video>	Generate subtitles for a video
subtitle models --list	List available Whisper models
subtitle models --download <model>	Download a specific model
subtitle batch --input-dir <dir>	Batch process multiple videos
subtitle formats	Show supported output formats

Options

Option	Description
--model, -m	Model to use: tiny, base, small, medium, large
--format, -f	Output format: vtt, srt, txt, json, lrc, ass, ttml
--merge	Embed subtitles into the video file
--threads, -t	Number of processing threads
--verbose, -v	Enable verbose output

Python API

from subtitle_generator.core import SubtitleGenerator, WhisperCppTranscriber
from subtitle_generator.models import ModelManager

transcriber = WhisperCppTranscriber(binary_path="./binary/whisper-cli")
generator = SubtitleGenerator(transcriber=transcriber, model_manager=ModelManager())

result = generator.generate(
    input_path="video.mp4",
    model_name="base",
    output_format="srt",
    output_dir="data",
)
print(f"Subtitles saved to: {result.output_path}")

Models

Model	Size	Speed	Accuracy
tiny	~75MB	⚡⚡⚡⚡	⭐⭐
base	~140MB	⚡⚡⚡	⭐⭐⭐
small	~460MB	⚡⚡	⭐⭐⭐⭐
medium	~1.5GB	⚡	⭐⭐⭐⭐⭐
large	~3GB	🐢	⭐⭐⭐⭐⭐

Tip: Use .en models (e.g., base.en) for English-only content for faster processing.

Links

• 📖 Documentation
• 🐛 Issue Tracker
• 📦 Source Code

License

MIT License - see LICENSE for details.