Home
cd ../playbooks
Media ProductionBeginner

AutoSubs: Local AI Subtitles and Transcripts With Speaker Labels

Transcribe audio and video on your own machine with AutoSubs, and let Claude run its command line to produce SRT, VTT, readable transcripts and word-timed JSON, with optional speaker labels and translation

15 minutes
By tmoroneySource
#subtitles#transcription#whisper#captions#speaker-diarization#srt#davinci-resolve#local-ai

Cloud transcription tools charge by the minute and want your recordings on their servers. AutoSubs runs Whisper and other speech models on your own machine, and its command-line mode means Claude can transcribe a folder of interviews, label who's speaking, and hand you subtitle files without you opening an app.

Who it's for: video editors and YouTubers adding captions, journalists and researchers transcribing interviews, podcasters, course creators, teams handling recordings that shouldn't leave their machines

Example

"Transcribe interview.mp4 with speaker labels, two speakers, English, and give me an SRT" → autosubs interview.mp4 --diarize --max-speakers 2 --lang en -o interview.srt, a model chosen to fit your RAM, and a spot-check of names and jargon

CLAUDE.md Template

New here? 3-minute setup guide → | Already set up? Copy the template below.

Or install from your terminal:
npx ccplaybooks add autosubs-local-subtitles
# AutoSubs: Local AI Subtitles and Transcripts From the Command Line

Use this when someone wants subtitles or a transcript from audio or video without sending the file to a cloud service: an SRT or VTT file, a readable transcript with speaker labels, a JSON file with word-level timestamps, or a translation to English. Trigger phrases: "transcribe this video", "make subtitles", "generate an SRT", "who said what", "add captions".

Source: https://github.com/tmoroney/auto-subs (MIT, by tmoroney). AutoSubs is a desktop app (standalone, or connected to DaVinci Resolve, Adobe Premiere Pro and After Effects) that also has a command-line mode built for terminals and AI agents. The project states that transcription runs on-device with no cloud, no subscription, and no data leaving the machine. This template drives the **command-line mode**.

## Install (let the user do this part)

The user installs the app from the project's official releases page (https://github.com/tmoroney/auto-subs/releases/latest). Installers are provided for Windows (`.exe`), macOS Apple Silicon and Intel (`.pkg`) and Linux (`.deb`, `.rpm`). macOS can also use Homebrew: `brew install --cask auto-subs`.

Debian or Ubuntu:

Get new playbooks like this one

One email a week with new Claude Code workflows. Free, like everything here.

No spam. Unsubscribe anytime.

README.md

What This Does

AutoSubs is a free, open-source desktop app for AI subtitles that the project says runs entirely on your own device, with no cloud and no subscription. It also has a headless command-line mode, so Claude can run it for you. It can:

  • Turn audio or video into timestamped subtitles and transcripts
  • Label speakers (diarization) so each person can be styled or quoted separately
  • Transcribe or translate across a large range of languages (the project lists 1,000+)
  • Export SRT, VTT, readable text, or JSON with word-level timestamps
  • Work as a standalone app, or connect to DaVinci Resolve, Adobe Premiere Pro and After Effects

It supports Whisper (several sizes), Moonshine, Parakeet, SenseVoice, Canary, Cohere Transcribe, GigaAM and Omni-ASR models, downloaded on demand. This playbook covers the command line.


Quick Start

Step 1: Install AutoSubs

Download the installer for your system from the project's official releases page (https://github.com/tmoroney/auto-subs/releases/latest), or on macOS use Homebrew:

brew install --cask auto-subs

On Debian or Ubuntu:

wget https://github.com/tmoroney/auto-subs/releases/latest/download/AutoSubs-linux-x86_64.deb
sudo apt install ./AutoSubs-linux-x86_64.deb

To use the autosubs command on macOS or Windows, open the app, go to Settings, then Command line, and click Install. On Linux it's already on the PATH.

Step 2: Create a Project Folder

mkdir -p ~/Documents/Transcripts

Step 3: Download the Template

Click Download above, then:

mv ~/Downloads/CLAUDE.md ~/Documents/Transcripts/

Step 4: Start Working

cd ~/Documents/Transcripts
claude

Try: "Transcribe interview.mp4 with the small model and give me an SRT." Claude will tell you the model's size before the first download.


Examples of What to Ask

You say Command Claude runs
"Give me a readable transcript" autosubs file.mp4 --model small
"Who said what? Two speakers" autosubs file.mp4 --diarize --max-speakers 2 --lang en
"Make subtitles with at most two short lines" autosubs file.mp4 -f srt --max-lines 2 --density less
"I need word-level timestamps" autosubs file.mp4 -f json
"Translate to English" autosubs file.mp4 --translate
"All caps, no punctuation" --text-case uppercase --remove-punctuation
"What models are available?" autosubs --list-models

Which Model?

Need Model Approx. size
Quick drafts, English moonshine-tiny, Whisper tiny.en 60 to 80 MB
Everyday balance Whisper small 480 MB
Better accuracy Whisper medium, large-v3-turbo 1.5 to 1.6 GB
Best Whisper accuracy Whisper large-v3 3.1 GB (about 10 GB RAM)
Fast and accurate European languages parakeet 700 MB
Translation to English canary 1 GB
Rare languages omni-asr-1b-ctc 3.7 GB

Sizes are approximate and the first run downloads the model, so pick one that fits your RAM.


Tips & Best Practices

  • Test on a short clip first. Check the language, model and subtitle formatting before a two-hour file.
  • Give it your vocabulary. The --prompt option guides transcription, so list names, products and jargon that are often misheard.
  • Spot-check the output. Check the start, the middle and every name. Speech recognition makes mistakes, especially with accents, noise and overlapping speakers.
  • Let the extension choose the format. -o subs.srt or -o transcript.json is enough.
  • Pipe-safe output. The transcript goes to stdout and progress to stderr, so autosubs file.mp4 -f srt > out.srt stays clean.
  • Use a GPU if you have one. It's auto-detected, and --gpu or --no-gpu forces it.

Limitations

  • Big downloads and big RAM. The best models are several gigabytes and need up to about 10 GB of RAM.
  • Accuracy varies by language. The project's ratings are relative and approximate. Rare languages and noisy audio need checking.
  • Word-level alignment for some languages is degraded. It relies on romanising the text, so very low-resource scripts may get weak or missing word timestamps.
  • Mac App Store DaVinci Resolve isn't supported for the Resolve integration.

Safety Notes

An automated security scan of the whole repository returned thousands of findings. Nearly all are in the project's installers, build and CI files, test files and dependency lock files, which are what you'd expect for a cross-platform desktop app. The points below are the ones that matter when you use it. A static scan isn't proof of safety, and this one only partially covered the repository.

  • Installers run with elevated privileges. The Windows, macOS and Linux installers run scripts at install time, and the Linux route uses sudo. Download only from the official releases page or the Homebrew cask, and read a command before approving it.
  • The macOS command-line install asks for your password. Clicking Install in Settings creates a link in /usr/local/bin. Do it yourself, and use Remove to undo it.
  • Resolve and Adobe integrations add scripts or extensions to other applications, and the Adobe extension loads automatically. Set those up only if you want them.
  • Check licences for commercial use. The code is MIT, but the optional MMS word-alignment model is CC BY-NC 4.0 (non-commercial), and every other model has its own licence.
  • Recordings of other people. Transcribing interviews and calls can carry privacy and consent obligations, even when processing stays on your machine.
  • The "nothing leaves your machine" claim is the project's own. Models are downloaded from the internet on first use. If it matters for sensitive audio, check the app's network behaviour yourself.

$Related Playbooks

Media Production

ASR Transcribe to Text

Transcribe audio and video to text with Qwen3-ASR — local MLX inference on Apple Silicon (no API key, 15–27x realtime) or a remote API endpoint, auto-detecting the best path.

10 minutes
Intermediate
Media Production

Claude Video: Let Claude Watch Any Video

Turns a video URL or local file into frames plus a timestamped transcript, so Claude answers from what it saw and heard instead of the title

10 minutes
Intermediate
Media Production

Editly: Make Videos From Code With Node.js and ffmpeg

Describe a video in a JSON5 spec (clips, images, titles, transitions, music) and let Claude write it and render it with Editly, for slideshows, promos, GIFs and batch resizing

10 minutes
Intermediate
Media Production

Creator Style Reverse-Engineer: One Video In, Style Guide Out

Feed Claude a video from a creator you admire and get a breakdown of their style, a transcript, and a step-by-step replication plan, then blend several creators into your own style

10 minutes
Intermediate
Media Production

Herk Edit Kit: AI Video Editing for Talking-Head Footage

Clone a video-editing kit and let Claude Code cut your silences, review your mistakes, plan the story and build motion graphics, with you approving every cut

30 minutes
Advanced
Media Production

Panel Moderator Guide

Generate a complete moderator guide - sharp questions, minute-by-minute agenda, follow-ups, and audience prompts for any panel.

5 minutes
Intermediate
Media Production

Podcast Production Team

A coordinated five-role agent team — researcher, scriptwriter, show-note editor, distribution manager, and reviewer — that produces a full podcast episode's content in one pass, with the team size scaling to match the actual request.

10 minutes
Intermediate
Media Production

Podcast Production Assistant: Show Notes, Mastering and Distribution

A Claude Code workflow for podcast post-production: master audio to -16 LUFS, generate chapters and timestamped show notes from a transcript, draft promo posts, and run a distribution checklist for every episode.

10 minutes
Advanced
Media Production

Podcast Creator System

Full podcast production workflow from episode ideation and guest research to show notes, SEO optimization, and promotional content.

10 minutes
Intermediate
Media Production

Spotify Integration & Analytics

Automate Spotify music playback, playlist management, and audio analysis workflows

10 minutes
Advanced
Media Production

Social Media Clip Creator

Repurpose long-form video and audio into short-form clips with optimized captions, hooks, and hashtags for TikTok, Reels, and Shorts.

10 minutes
Beginner
Media Production

Transcript Fixer

Correct speech-to-text errors with dictionary rules and AI analysis — fixing homophones, garbled technical terms, and mixed Chinese/English — while building a personalized correction database that learns from each fix.

10 minutes
Intermediate

Browse all Media Production playbooks →