Speaking

Speech models and language models

Two models, two different jobs. On the official cloud relay both are already set up for you — this page is for talking over the same Wi-Fi, or for a server you run yourself.

The speech model — it turns your voice into words

On the PC: FlowMic → Settings → Speech recognition → “Built-in speech model”. Pick a language there and download it. Nothing downloads until you ask for it.

Download the model for the language you actually speak. A model for another language still produces words — the wrong ones.

The same card has a “Download folder” you can change, which is worth doing when your system drive is short of space.

Changing that folder does not move what you already downloaded. Move those files yourself, or download them again into the new folder.

If you would rather use another recognizer

The same page has an Engine preset menu, grouped the same way as the language-model one: the built-in offline engine first, then recognizers you run yourself, then cloud services you have your own account with, and a Custom entry for anything else. You only need this if you do not want the built-in offline engine — it is what a fresh install already uses.

The language model — it rewrites what you said

On the PC: FlowMic → Settings → Language model. Open the Model preset menu and pick the service you already use — it fills in the protocol, the address and a sensible model name for you, so there is nothing to type from memory. On a fresh install the menu reads “Please choose”, because nothing is set up yet.

The menu on your PC, grouped. Vendor names are the same in every language; the list grows with releases, so trust the one in your app.

Cloud services — you bring your own key

Pick your provider and the address and model name arrive with it. The key stays yours: you sign up with them, you pay them, and you paste the key into the API Key field. FlowMic has no account with any of these and never supplies a key.

Self-hosted — the model runs on your own machine

Picking one of these fills in a local address as a starting point. Change it if your server listens somewhere else, and set the model name to whichever model you actually loaded. A server on your own machine usually needs no key at all.

Custom — anything else that speaks the same protocol

Pick Custom and type the address yourself. Editing any field by hand also switches the menu to Custom, so the name at the top never claims a service you have edited away from.

Then tap “Test connection”. It checks the address, the key and the model on the spot, and tells you which of the three it did not like.

The key comes from whoever runs that model — your account with a provider, or your own server if you host it yourself. FlowMic never supplies one.

A rejected key and a model that is not there are reported separately, by name, so you can tell which of the two you are looking at.

What stops working when there is no language model

• Translate and Organize: neither mode can run. Realtime is unaffected — it never needed one.
• AI polish: the switch stays where it is and nothing happens.
• Your scenario card: the terms in it keep working. They are handed to the speech model as hints, which needs no language model at all — only the rules for AI polish are skipped.