SRT-Maker

Web app to generate Premiere-ready SRT files from audio and video — self-hostable, runs Whisper locally with CUDA.

Features

• Transcribes audio (MP3, WAV, M4A, OGG, FLAC) and video (MP4, MKV, WebM, MOV) files
• Uses faster-whisper large-v3 with GPU acceleration
• Pause-aware subtitle grouping — breaks cues at natural speech pauses and punctuation
• Gap-free timing — no loose frames between consecutive subtitle cues
• Drag-and-drop web UI with dark mode
• Automatic SRT file download

How it works

1. Upload an audio or video file via the web UI
2. Audio is extracted from video files using FFmpeg
3. faster-whisper transcribes with word-level timestamps
4. Words are grouped into subtitle cues based on:
   • Natural speech pauses (>=300ms gaps)
   • Punctuation (sentences, commas, semicolons)
   • Dynamic word count (2-5 words depending on speech rhythm)
5. Gaps between cues are closed (up to 500ms) to eliminate loose frames
6. SRT file is returned for download

Setup

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Requires FFmpeg for video file support.

Usage

python main.py
# or
uvicorn main:app --host 0.0.0.0 --port 8765

Then open http://localhost:8765 in your browser.

Deployment

The app runs as a systemd service behind a Cloudflare Tunnel at srt.lukanet.work.