Live Translation

Real-time captions and translations from microphone, system audio, or both — with an optional always-on-top overlay window so the captions sit over whatever you're watching.

What you can do with it

• Live meeting captions — caption a Zoom / Meet / Teams call in another language without joining as a translator bot.
• Real-time language learning — caption foreign-language content (films, podcasts, lectures) with your native language as the translation track.
• System-wide subtitles — capture system audio so you can subtitle YouTube / Netflix / anything that plays on your speakers.

What you need

• FFmpeg on PATH — see FFmpeg setup.

• An STT backend, one of:

  • faster-whisper — local, offline, free, default
  • Soniox — cloud, paid, real-time speaker diarization. See Soniox setup.

• For system audio capture, the right backend per OS is auto-selected: Linux uses parec (PulseAudio / PipeWire), Windows uses native WASAPI loopback (no extra software in most cases), macOS uses ffmpeg -f avfoundation against a virtual loopback device (BlackHole / Loopback / etc.). An inline warning banner with clickable install links shows up if anything's missing. See Setup → System audio for full per-OS install instructions.

Walkthrough

1. Click Live Translation in the sidebar.

2. Configure once in Settings → Live:

   • Source language (spoken language)
   • Target language (or leave blank for transcription only)
   • Audio source: Microphone / System audio / Both
   • STT method: Whisper / Soniox

3. Back on the Live page, click Start (Ctrl+Enter).

4. The transcript fills the main pane card-by-card. The floating Overlay window also shows captions (drag it to wherever you want).
5. Click Stop to end the session.

The transcript view

Pick a layout in the toolbar:

• Both stacked — original + translation, one above the other
• Both side-by-side — original on the left, translation on the right
• Original only / Translation only

The toolbar buttons use ON / OFF suffixes for at-a-glance state — e.g. TTS ON, TTS OFF, Timestamps ON, Overlay OFF.

Toggle timestamps on/off with the clock icon. Toggle TTS playback of the translated lines with the speaker icon. Honours your Settings → Voice → TTS method pick — Edge TTS (default), ElevenLabs, Google Cloud TTS, Gemini TTS, or Piper TTS (fully offline). With Piper selected, missing per-language voices silently fall back to Edge TTS mid-stream — there's no modal pre-flight on this page, since blocking the live flow on a download dialogue would be worse than the fallback.

The overlay window

A draggable, resizable, always-on-top tool window. Shortcuts:

Shortcut	Action
Ctrl+[ / Ctrl+]	Decrease / increase opacity
Ctrl+Arrow	Move the overlay
Ctrl+0 / Ctrl+9	Grow / shrink

Position, size, opacity, and font size persist between sessions.

Live-sync with Settings

Font size and opacity controls work in both directions: dragging the Font size or Opacity slider in Settings → Live → Overlay Configuration updates the open overlay in real time, and conversely, pressing + / - / Ctrl+[ / Ctrl+] inside the overlay updates the sliders in Settings. No overlay restart required.

Empty-state placeholder

Before any audio is captured the overlay shows a placeholder ("Press Start..." idle / "Listening..." once Start is clicked) that mirrors the main window's empty state — the swap stays in lockstep with the running status pill. The placeholder scales with the overlay's current width × height so it remains readable at any window size.

Minimal-captions mode

The Show minimal captions checkbox in Settings → Live → Overlay Configuration hides the timestamp + speaker chips on the overlay while keeping them visible on the main window. Useful when the overlay is shared with an audience (presenter mode / screen sharing) but you still want full metadata in your working view. The toggle is overlay-only — it doesn't change your "Speaker labels" preference for the main window.

Save the transcript

Click Save Transcript to export the session to a .txt file with timestamps, speakers, original lines, and translated lines.

Picking an STT backend

Backend	Best for	Cost	Latency
Whisper (local)	Offline, privacy-sensitive	Free	Medium (~1 s after end-of-sentence)
Soniox	Multi-speaker meetings	Paid (~$0.005 / min)	Low (real-time)

Caveats

Microphone selection

The mic input always uses the OS default device — there's no in-app picker (sounddevice surfaces too many ALSA virtual plugins to be useful, and the OS already owns the default-mic UI). Set your preferred mic in your OS sound settings before starting.

TTS backpressure

The TTS queue is bounded to the most recent 3 sentences — older queued audio is dropped if synthesis falls behind. This keeps spoken playback near the on-screen captions.

ElevenLabs without a key

If you've set TTS method to ElevenLabs but no API key is configured, the Live page automatically falls back to Edge TTS and announces the fallback in the status label.

Shortcuts

Shortcut	Action
Ctrl+Enter	Start / Stop
Ctrl+K	Clear log (with confirmation)
Ctrl+[ / Ctrl+]	Adjust overlay opacity