Live transcription
Transcription turns the interviewer's audio into text you can read while they are still talking.
How it works
Once a session starts, a rolling caption shows what is being said. Transcription runs in the cloud on every account, so there is no speech engine to install or choose. It helps when you miss a word, and when a question runs long enough that the beginning has already slipped.
Languages
Cloud transcription covers 60+ transcription languages.
Pro's 3,000 credits cover more than 40 live hours on GPT-6 Astra, GPT-6 Luna or Grok 4.7; other models use credits at different rates.
Audio sources
| Source | What it captures |
|---|---|
| Speaker | Audio playing on your machine — the interviewer's voice. |
| Microphone | Your own voice. |
Choose the speaker the interviewer comes through and the microphone you use. Those follow your computer's current devices, and they update when the devices change. Pick a specific one when the call is on a headset or a virtual audio cable.
Mock interviews listen only to your microphone — you are the candidate. Live copilot starts on the interviewer's audio, with your microphone muted. Unmute it when you want your side in the caption too.
Sending a question to the assistant
- ⌘⌥↵ — send the live caption, or what you typed, plus any attached screenshots.
- ⌘⌥, — clear the draft text without removing attached screenshots.
Send includes the interviewer’s latest caption, so you do not copy it into the draft first. Attach a screenshot, then send, when the question is on screen.
If detection feels off
Interviewers speak at very different paces. Endpoint timing is set on the server for cloud transcription — there is no pause-interval control in Settings. If captions cut early or run long, check the selected speaker and microphone under audio sources first.