How Interview Signal records interviews without a bot

Interview Signal never joins the meeting. It takes two streams of audio on the interviewer's own computer: the microphone, labeled "You", and whatever the computer plays, which during a call is the candidate, labeled "Candidate". Because the label comes from which device the sound came from, no bot, meeting integration or voice matching is needed, and it works the same in Zoom, Microsoft Teams, Google Meet or a phone screen taken through the computer. The audio goes in short chunks to a speech-to-text service and comes back as text; no audio file is saved. This describes version 0.9.10, the current release as of October 2026.

How does Interview Signal hear both sides of the call?

It runs two separate recorders, called lanes:

The pre-interview screen says it in one line: "Your microphone is labelled as you; the meeting audio is labelled as the candidate." Keeping the lanes apart is what makes the labels reliable on a video call or phone screen, where the candidate is the only other voice.

The system-audio lane uses the operating system's screen-and-audio capture path, which insists on naming a screen even when only sound is wanted. Interview Signal names one and stops the video the moment it arrives; nothing from the screen is recorded. On a Mac this is why the app asks for the Screen & System Audio Recording permission, and why you have to quit and reopen it after allowing it. Windows asks only for the microphone.

Does the candidate see anything on the call?

No. There is no bot, no extra participant and no recording banner from a third party, because nothing is added to the meeting at all. The app listens on your computer only. That means the candidate learns about it only if you tell them, which is what the read-aloud line in the consent features is for.

What happens to the audio?

1. Each lane records continuously while the interview runs.
2. Every few seconds, the app sends the latest stretch of each lane, through Interview Signal's server, to a speech-to-text provider. The security page names it: Groq, with OpenAI as the fallback.
3. Text comes back and shows as a live caption under You or Candidate.
4. When the speaker pauses, the words become part of the transcript, which is saved on your computer every few seconds. The guide, follow-ups and scorecard update from it.
5. The audio is processed and discarded. It is never written to a file on your computer, and it is not retained by the provider or by Interview Signal.

Audio that is near silence is not sent. Speech recognizers tend to invent stock phrases over silence, such as "thank you for watching", and not sending silence is how the app keeps those out of the candidate's mouth. Where the transcript and everything else is kept is covered in where your data is stored.

How does it work for phone screens and in-person interviews?

The pre-interview screen asks "How are you speaking?", with three answers:

• Video call: the normal case. The meeting app plays the candidate through your computer.
• Phone: works the same way, as long as the call runs through the computer, for example in a calling app on it. A call on a separate handset never reaches the computer's sound output, so it cannot be captured.
• In person: the app warns, "In a room, both voices reach the same microphone, so speaker labels will not be reliable. The transcript and scorecard still work." Pressing "Mute me" in person points you to "Pause capture" instead, because your microphone is carrying the candidate too.

What if only one side is captured?

If the computer's sound cannot be captured, the interview still runs on your microphone alone, with a banner that stays up: "System audio is not available — recording your microphone only. The candidate will not be transcribed." Your questions are still transcribed and the guide still ticks off. If system audio stops partway through, for example because the meeting app released it, the banner reads "System audio stopped — still recording your microphone. Restart the interview to capture the candidate again." Fixes for both are in audio troubleshooting.

What happens if I use speakers instead of headphones?

Your microphone hears the candidate through the speakers, so both lanes pick up the same sentence. The app matches the duplicate against what the candidate said and removes it from the transcript, and the first time it happens it shows a note: "Your microphone is picking up the candidate through your speakers. The duplicate is being removed — headphones would give a cleaner transcript." Headphones stop it at the source.

What is the difference between Mute me and Pause capture?

• "Mute me" (Ctrl+M, or Cmd+M on a Mac) stops your lane only, for an aside or a cough. The banner says "You are muted — the candidate is still being transcribed and scored."
• "Pause capture" stops both lanes. Nothing is buffered or sent, and the banner "Paused — nothing is being recorded." stays until you press "Resume capture".

What are the limits of capturing audio this way?

• Everything the computer plays counts as the candidate. A video in another tab or a notification with speech is transcribed under Candidate, so mute other sound before you start.
• Two people on one microphone cannot be told apart, which is the in-person case above.
• There is no recording to replay. No audio or video file is kept, only the text transcript.
• It needs a connection for transcription during the call.

For the general idea behind this approach, and how it differs from a meeting bot, see what a bot-free AI note taker is. Before your first interview, run the audio check to confirm both lanes are heard.