Turns what you say into text on your own server, so Home Assistant's voice assistant understands you without a cloud
On this page
Good to know before you install
It has no page of its own: Home Assistant on this server uses it.
At a glance#
- What it is
- Speech to Text AI
- Where it opens
- No web page It works in the background
- Updates
- Automatic A backup is taken first, and it rolls back if the update fails
- Backups
- Left out Not in the scheduled backups unless you switch it on
- Copies
- One per server
There's no step-by-step guide for Whisper here yet. Everything below comes straight from the app catalogue, so it's what LibrePortal really does when you install it. For using Whisper itself, its own documentation is the place to go: github.com/OHF-Voice/wyoming-faster-whisper.
What the install asks#
When you install Whisper, LibrePortal asks for a little more than usual before it starts:
- Speech model
Addresses and ports#
Whisper has one port. LibrePortal picks a free port number for each when you install, so yours will differ from anyone else's. The Config tab's Network section shows them, and the app's Open button uses the right one.
| What it is | Who can reach it | With a domain |
|---|---|---|
| Speech to text (Wyoming) TCP | local | Port only |
Settings#
These are Whisper's own settings on its Config tab. The settings every app shares, like backups, updates and where its data lives, are explained in App pages and settings.
| Setting | What it does | Starts as |
|---|---|---|
| Speech Model | Bigger models make fewer mistakes and answer more slowly. Choices: Fast (tiny), Balanced (base), Accurate (small) |
base-int8 |
| Language | Used when the app asking does not say; Home Assistant does. Choices: This server's language, English, German, French, Spanish, Italian, Dutch, Portuguese, Polish, Swedish, Danish, Norwegian, Finnish, Czech, Russian, Ukrainian, Turkish, Greek, Hungarian, Romanian, Catalan, Welsh, Vietnamese, Chinese |
auto |
In its config file#
A few more aren't on the Config tab yet. They live in the app's config file, described here so you know they exist. How to change them.
| Setting | What it does | Starts as |
|---|---|---|
NameCFG_WHISPER_DISPLAY_NAME |
What this app is called on your dashboards. Leave empty to use its usual name | Empty |
Show on the LibrePortal DashboardCFG_WHISPER_SHOW_ON_DASHBOARD |
Whether this app gets a shortcut on the front page of the LibrePortal WebUI, separate from any Dashy dashboard Choices: true, false |
false |
Model Update CheckCFG_WHISPER_UPDATE_CHECK |
Whether Whisper may contact Hugging Face again after its model is downloaded Choices: true, false |
false |
Works with#
These are the other apps Whisper knows how to work with. Once it's installed, its page shows the same list with whether each one is connected. How this works.
- Grafana: Shows dashboards for the apps that have one. works with every app
- Home Assistant: Understands and answers you without a cloud.
- Prometheus: Collects metrics from the apps that can export them. works with every app
- Uptime Kuma: Watches your apps and shows when one goes down. works with every app
Privacy and security#
All clear Whisper turns speech into text on your own server. It downloads its language model once from Hugging Face and then works without the internet.
What leaves this server#
- The speech model is downloaded once from Hugging Face
What LibrePortal turns off#
- Hugging Face telemetry (switchable in Config)
Licence: MIT (wyoming-faster-whisper; verify). Audited: 21 October 2026. See the privacy of every app and the Promise.
Sources#
Under the hood#
- It runs this image:
rhasspy/wyoming-whisper:3.8.1@sha256:ba6fcb6056ebe237d15a325381763a80af2fc8fcedaa04f6a714a7375eb20d80. - It can sit behind Authelia's sign-in once Authelia is installed.
- Its own project and documentation: github.com/OHF-Voice/wyoming-faster-whisper
Ports, settings, tools and connections on this page are read from LibrePortal v0.10.97's app catalogue each time the guide is built.