-
Notifications
You must be signed in to change notification settings - Fork 0
Glossary
Plain-language definitions for terms used across the Voxis Live app, README, and Discussions.
"Bring Your Own Key." In the open-source build, you supply your own Gemini API key instead of signing in to a hosted service. The key is stored encrypted on your machine and used only to talk directly to Google's API — no Voxis server is involved.
The engine behind the app's free daily minutes. Instead of a fully voiced cloud model, it uses the same translation model for text and speaks the result with a local, on-device voice. It costs a fraction of the paid engine, which is why it's what free usage runs on after a short full-quality trial.
The default audio capture path. It reads the Windows system audio mix directly and excludes the app's own output, so no virtual audio cable needs to be installed for one-way translation.
Automatically lowering the volume of the original audio while the translated speech plays, so the two don't talk over each other.
The delay between hearing something in the source language and the translation being produced. Human simultaneous interpreters have one too — it's the time needed to gather enough context to translate accurately. It's a property of simultaneous interpretation itself, not something the client can tune away.
Two-way translation for live calls: you hear the other party translated into your language, and your own speech is translated and sent out through a virtual microphone so the other party hears it in their language.
A translation model that takes a continuous audio stream and produces continuous translated audio, self-balancing accuracy against how far behind it stays — as opposed to waiting for a complete sentence, transcribing it, translating the text, then synthesizing speech as three separate steps.
A Windows audio capture technique that reads the system's audio mix while excluding one specific process's own output. Voxis uses this to capture "everything except its own translated voice," which is what makes driverless mode possible without a feedback loop.
Named latency/quality trade-off profiles (Smooth, Fast, Callout, Saver) rather than a continuous slider — see Feature Reference.
Local, on-device detection of when the speaker changes during a session, used to tag captions and transcripts with anonymous labels (S1, S2, …). It detects a change in voice, not who the person is.
One-way translation for anything you're watching or playing: the original audio is ducked and the translation plays over it.
A virtual audio device that lets one application's output be selected as another application's microphone input. Voxis uses this to feed translated speech into a meeting app as if it were a real microphone, for the outgoing leg of Meeting mode.
Windows Audio Session API — the low-level Windows audio interface Voxis uses for both capturing system audio and controlling per-application volume (used for ducking).