Live captions for your Discord call, straight into OBS.

Free

No tiers, no licence key, no account, no API key, nothing metered. It’s MIT licensed, so fork it and ship your own if you want to.

It runs on your PC

Your voices never leave the machine. ChatterLayer downloads a speech model once and does the listening itself, so after that it doesn’t need the internet to caption at all. The caption server answers on 127.0.0.1 and nowhere else, unless you deliberately switch sharing on. The app makes one request of its own accord, a version check at launch, and there’s a switch for that.

Sharing, if you want it

Co-streaming? One person runs the app and sends the others a link to paste into OBS, and everyone shows the same captions. It’s off until you turn it on, it never starts by itself, and it stops when you close the app. Only the finished text goes down the tunnel. The audio stays on your PC.

Nine speech models

Four engines, nine models, 40 MB to 2.5 GB. The small one keeps up with a seven-person call. The big one gets close to what you’d pay a cloud service for. Both of the recommended ones punctuate. Keep several installed and switch between them without downloading anything again.

Nobody by default

Everyone in the call starts switched off. You turn on the people who’ve said yes, and you can change that mid-call without reconnecting.

Light on your PC

It’s a 30 MB download and takes about 95 MB once installed. Waiting in the tray, it uses next to no CPU. While it captions, nearly all the work is the speech model, and since 0.5 that runs as native code: on Parakeet it uses about 15% less CPU than 0.4 did, and the first caption arrives about 4 seconds sooner.

It’s just a browser source

The overlay is an ordinary OBS browser source with a transparent background. No plugin, no window capture, nothing to configure. It reconnects on its own, so you can add it before the app is even open.

It started with one viewer.

Three hops, and the middle one is the only place your voices are ever decoded.

1. Discord

   A bot sits in your call

   ChatterLayer brings a bot into your voice channel, and Discord hands it every person’s audio as a separate stream. Everyone gets transcribed on their own, so two people talking at once never get spliced into one sentence.

2. Your PC

   Your PC does the listening

   Discord’s audio is decoded straight to the format the speech engine wants and handed to it on its own thread. You choose the model, nine of them across four engines. It loads once and serves the whole call, so a seventh person doesn’t mean a seventh copy of it sitting in memory.

3. OBS

   Captions land on stream

   Every line goes out over a local WebSocket in that person’s colour. The phrase models show a faded guess first and rewrite the line once it settles. Vosk writes it word by word as they speak.

In Both, press “This is me” on your own Discord row. Your mic captions you from then on, so you never appear twice.

Measured on a Ryzen 5 5600X.

About ten minutes, and most of that is a download bar.

1. 01

   Install it

   On Windows there’s an installer, or a zip if you’d rather not install anything. Macs get a .dmg, and Linux an AppImage or a .deb. The whole setup runs about ten minutes, and most of that is watching a download bar.

   Get the download

   None of the builds are code-signed, so Windows SmartScreen throws up a blue box the first time you run it: click More info, then Run anyway. On macOS, Gatekeeper may call the app damaged; one Terminal command clears that, and it’s shown right after you download.

2. 02

   Pick a speech model

   First launch asks which model to download. There are nine, across four engines, and they run from 40 MB to 2.5 GB. You only need one, which is why none of them ship with the app. ChatterLayer checks your RAM and CPU and marks the one that suits your machine: Parakeet TDT 0.6B if you have 16 GB or more, which most streaming PCs do, and Moonshine Base on lighter ones.

   Install as many as you like and switch between them in the dropdown without downloading anything again. They live in your user data folder and survive app updates, and Remove gives you the disk space back. Switching to a different engine needs a reconnect.

3. 03

   Make a Discord bot

   The bot is the thing that sits in your voice channel and listens. Making one is free and takes about two minutes.

   • New Application, name it whatever you like, then the Bot tab and Add Bot.
   • Reset Token, then Copy. Treat it like a password.
   • Leave every Privileged Gateway Intent switched off. ChatterLayer doesn’t need them.

   Discord Developer Portal

   Only captioning your own mic? Skip this step and the next: pick My mic under Source and go straight to OBS.

4. 04

   Paste the token, pick a channel

   Drop the bot token into the Source panel and it signs in on its own. Press Invite the bot to a server, pick your server in the browser tab that opens, and it shows up in the Server dropdown by itself. Choose a voice channel and hit Connect. There’s no channel ID to copy and no Developer Mode to turn on, and channels your bot can’t join are greyed out with the missing permission named.

   The token goes into your operating system’s own keystore, DPAPI on Windows, Keychain on macOS, libsecret on Linux, so you paste it once and never think about it again.

5. 05

   Add the overlay to OBS

   Sources, then +, then Browser. Set the width and height to match your canvas, usually 1920 × 1080, and paste the URL from ChatterLayer’s Output panel:

   Copy

   • Tick Shutdown source when not visible and Refresh browser when scene becomes active.
   • Captions sit bottom-left with a small margin, so a full-canvas source at 0,0 usually just works.

   The overlay reconnects by itself. You can add the source before ChatterLayer is running, and it will survive you restarting the app mid-stream.

Stuck on any of it? Open an issue and say where it went wrong. There’s no support queue and no forum — it’s one person reading them.

Off until you switch it on. It never starts by itself, and it stops when you close the app.

In a co-op stream, only one person needs to run ChatterLayer. They turn on a share link and everyone else pastes it into a browser source, so the whole lineup shows the same captions with the same colours.

1. Tick the box

   Open Remote overlay and tick “Let other streamers use these captions”. Nothing has opened yet. That only reveals the controls.

2. Start the tunnel

   The first time, ChatterLayer fetches cloudflared, about 35 MB, and keeps it for next time. If you already have it on your PATH, it uses that copy instead.

3. Send the link

   Your co-streamers paste it into a browser source, exactly like the local one. Press Stop tunnel, or just close ChatterLayer, and the link stops working.

• The link changes every time you start the tunnel, because the hostname is random, so your co-streamers have to re-paste it after you restart ChatterLayer. Sort that out before you go live, not during.
• Cloudflare makes no uptime promise for these free quick tunnels. They need no account and no domain, and that’s the trade.
• The audio still never leaves your machine. What goes through the tunnel is the finished caption text, the same words already on your stream, and nothing else.
• Changing the overlay port stops the tunnel, since the old link would point at nothing. Start it again for a new one.

You need exactly one. The app marks the one that fits your PC, and you can change your mind later.

Word by wordVosk

Vosk updates the caption as each word is spoken, so text appears with almost no delay. The trade is a lowercase stream with no punctuation, and every person needs their own recogniser, so memory grows with the size of the call.

A phrase at a timeMoonshine, Whisper, Parakeet

These wait for someone to finish a thought, then hand back a punctuated, capitalised sentence a fraction of a second later. It reads like writing rather than a transcript, and one copy of the model serves everybody in the call.

Measured on a Ryzen 5 5600X against real speech. Speakers is the point at which captions start arriving late, not a hard limit. Vosk Large and Gigaspeech are extrapolated from model size.

Parakeet TDT 0.6B

The one to pick if your PC has 16 GB of RAM or more, which covers most streaming machines and not just the expensive ones. It’s the most accurate model here by a distance and still quick, about 290 ms a caption, because it transcribes the phrase itself rather than a padded window. The cost is a 2.5 GB download and around 2.6 GB of RAM, loaded once no matter how many people are on.

Moonshine Base

The one to pick on lighter machines, like an 8 GB laptop. It beats Vosk Medium everywhere except immediacy: more accurate, punctuated, capitalised, a caption in about 150 ms and a tenth of the CPU per second of speech. And because one shared model serves the whole call, it uses less memory than Vosk Medium once more than three people are talking.

Vosk Medium

Pick this if you want text appearing as the words are spoken instead of at the end of each phrase. That immediacy reads very differently on stream, and it’s the one thing Vosk still wins on.

The Whisper models

Hard to recommend over Moonshine. Every Whisper caption pays for a padded 30-second window whatever the phrase length, so Tiny is slower than its size suggests and Base takes about a second while being less accurate than Moonshine. They’re here because Whisper is the name people ask for, and because Small is genuinely accurate for one person if you don’t mind waiting two seconds.

An installer, or a zip if you'd rather not install anything.