One command to turn any video into text.
aits "https://www.youtube.com/watch?v=VIDEO_ID"| Feature | Description |
|---|---|
| Fast | ~30s to transcribe 10 min audio on Apple Silicon (20x realtime) |
| Accurate | OpenAI Whisper large-v3-turbo, supports 99 languages |
| Wide Support | YouTube, Bilibili, Twitter, TikTok, and 1000+ platforms |
| Free | Runs entirely local, no API costs, no usage limits |
| Private | Audio never leaves your machine, fully offline |
Online Platforms (via yt-dlp):
- Video: YouTube, Bilibili, Vimeo, Dailymotion, TikTok
- Social: Twitter/X, Instagram, Facebook, Reddit
- Live: Twitch, Kick (VODs)
- Music: SoundCloud, Bandcamp
- Podcast: Apple Podcasts, RSS Feeds, mp3 files
Local Files: mp3, wav, m4a, mp4, mkv, and more
Full list: yt-dlp supported sites
On Apple Silicon Mac (M1/M2/M3) with default model:
| Audio Length | Processing Time | Speed |
|---|---|---|
| 10 min | ~0.5 min | 20x |
| 60 min | ~3 min | 20x |
| 120 min | ~6 min | 20x |
Use --fast mode for even faster processing with slightly lower accuracy.
Note: First run downloads the model (~1.5GB), which takes a few minutes. Subsequent runs use the cached model.
pip install audio-transcriberThat's it. The tool auto-installs dependencies (ffmpeg, deno, mlx-whisper) as needed.
# Simple: just paste a URL
aits "https://www.youtube.com/watch?v=VIDEO_ID"
# Transcribe local file
aits audio.mp3
# Fast mode (trade accuracy for speed)
aits "URL" --fast
# Japanese video translated to English
aits "URL" --language ja --translate
# Output SRT subtitles (for video editors)
aits "URL" --srt| Option | Short | Description |
|---|---|---|
--model |
-m |
Model size (default: large-v3-turbo) |
--language |
-l |
Source language (e.g., en, ja, zh) |
--translate |
-t |
Translate to English |
--output |
-o |
Output directory (default: ./output) |
--fast |
Fast mode (turbo + beam_size=1) | |
--srt |
Also output SRT subtitle file | |
--keep-audio |
Keep downloaded audio file | |
--dry-run |
Show info only, don't download or transcribe |
| Model | Speed | Accuracy | Recommended For |
|---|---|---|---|
large-v3-turbo |
★★★★ | ★★★★★ | Default, best speed/quality balance |
distil-large-v3 |
★★★★ | ★★★★ | Fast high-quality alternative |
medium |
★★★ | ★★★★ | Lower memory usage |
small |
★★★★ | ★★★ | Quick drafts |
base |
★★★★★ | ★★ | Very fast, lower quality |
tiny |
★★★★★ | ★ | Testing only |
First run downloads the model (~1.5GB), then uses cache.
- Python 3.9+
- FFmpeg (audio processing)
- Deno (required for YouTube downloads)
| Platform | Backend | Notes |
|---|---|---|
| Apple Silicon Mac | mlx-whisper | GPU accelerated, fastest |
| Intel Mac | faster-whisper | CPU |
| Linux | faster-whisper | CUDA GPU supported |
| Windows | faster-whisper | CUDA GPU supported |
- txt: Plain text (default)
- srt: SubRip subtitles, works with Premiere, Final Cut, DaVinci Resolve, etc.
python -m audio_transcriber strip-srt subtitle.srt