Call Translation
Your voice, translated in real time on Teams, Zoom or Meet
Call Translation turns your speech into a translated voice that the other person hears during the call. It is speech to speech: nobody has to read captions. You speak your language and the other side hears you in theirs — any of the 99 supported languages.
What you need
- An OpenAI API key configured in Speakly
- A virtual audio device installed (BlackHole on macOS, VB-CABLE on Windows) — both free
- Your calling app: Teams, Zoom, Meet or any other
Step 1: install the virtual audio device
The operating system has no way to hand one app's audio to another app as a microphone. The virtual device is that cable: Speakly speaks into it, and Teams listens to it as if it were an ordinary microphone. One-time setup, about five minutes.
macOS
With Homebrew, run the command below — or download the installer from existential.audio/blackhole. Either way macOS asks for your password, because it is installing an audio device. No restart needed, but quit and reopen Teams/Zoom afterwards so they pick up the new device.
brew install --cask blackhole-2chWindows
- Download VB-CABLE from vb-audio.com/Cable
- Right-click VBCABLE_Setup_x64.exe and choose Run as administrator
- Click Install Driver, then reboot
- It installs one pair of devices: "CABLE Input" (an output) and "CABLE Output" (an input)
Step 2: set up Speakly
- Open the Call Translation tab
- Leave "Your microphone" on your real mic (headset, AirPods) — that is what gets translated
- Set "Translated voice output" to the virtual device: "BlackHole 2ch" on macOS or "CABLE Input" on Windows
- Set "They hear" to the language the other person should hear
- Click "Start translating"
Step 3: set up Teams or Zoom
- Teams: … → Settings → Devices → Microphone. Zoom: Settings → Audio → Microphone
- Pick "BlackHole 2ch" (macOS) or "CABLE Output" (Windows) as the microphone
- Keep the Teams/Zoom speaker on your headphones so you still hear the other person
Translating what the other person says
Call Translation also works the other way round: you hear, in your own language, what the other person is saying. In this direction no virtual device is needed at all — Speakly captures the system audio directly.
- In the Call Translation tab, switch the input from "My voice" to "Their voice"
- Set "You hear" to your own language
- Keep the output on your headphones
- On macOS the first run asks for Screen Recording permission — that is how the system grants audio capture
What it costs
OpenAI bills this model per minute of session, not per word or per token: about $0.034 per minute for the translation and the voice (one model, they cannot be separated), plus about $0.003 per minute for transcribing what you say, which fills the text panel. In practice a 30-minute call costs around $1.11. Speakly shows a live estimate during the call, but your OpenAI invoice is the source of truth.
Common issues
- "No virtual audio device found": the device is not installed, or Speakly was opened before installing it — quit and reopen the app. Use the "Setup guide" button on the screen for the walkthrough.
- The other person hears nothing: check that the microphone selected in Teams/Zoom is the virtual device, not your real mic.
- You hear the translation but the call does not: Speakly's output is on your headphones instead of the virtual device.
- Echo or doubled voice: Speakly's output is pointing at a device that is also its input — never use the same virtual device as both microphone and output in Speakly.
- Region error: if your OpenAI organization has EU or US data residency, check the region configured in preferences.