Can an AI interview assistant transcribe both the interviewer and the candidate?

If you have tried to record an interview with a single microphone, you already know the problem: your own voice is loud and clear, but the interviewer — coming through your speakers or headphones — is faint, echoey, or missed entirely. Transcribing both sides cleanly needs two separate audio sources.

An AI interview assistant solves this by capturing two streams at once: system audio (everything your computer plays, including the interviewer's voice in Zoom, Google Meet, or Microsoft Teams) and your microphone (your own answers). Each stream is transcribed independently, so words are attributed to the right speaker instead of being blended into one muddy track.

SubcueAI runs as a native desktop app on macOS and Windows — not a browser plugin, and not a bot that joins the meeting. On macOS it taps system audio through the operating system's screen-capture audio APIs; on Windows it uses the system loopback device. Your microphone is captured in parallel through the normal input device.

Consider a backend engineer interviewing for a senior role over Zoom: the hiring manager's questions arrive as system audio, the candidate's spoken answers arrive as microphone audio, and SubcueAI transcribes each in real time. Because the two streams stay separate, the transcript reads as a labelled back-and-forth rather than one undifferentiated block.

The capture happens entirely on your machine, and the floating overlay that shows the transcript is local to your desktop. A short setup walkthrough covers the install.

Keeping the interviewer and candidate on separate channels does more than tidy the transcript — it lets the assistant decide which turn to act on. A question on the interviewer's channel is what an answer suggestion should respond to; your own channel is context, not a new prompt.

• Speaker labelling falls out naturally from the two-source design, since each stream has a known owner.
• Latency depends on the speech-to-text path; streaming transcription returns partial text while the person is still speaking.
• Accuracy is best with a clean system-audio signal — a good headset and a quiet room help both channels.

For the full capture-to-suggestion path, see the how it works hub.

Two-sided transcription assumes the audio actually reaches your computer's output and input devices. It does not cover every situation, and it is honest to say so:

• On a company-managed device you do not control, you may not be able to install a desktop app at all.
• Proctored environments and remote-monitoring software can restrict background apps or screen activity.
• If a session is screen-recorded on the other side, that recording is independent of anything your assistant does locally.
• In-person rounds, where audio never passes through your computer, are out of scope for system-audio capture.

SubcueAI is built for your own remote interviews on your own machine; its limits are covered on the detectability hub.