Features

Real-time voice AI interviews

Intervieux interviews run as a live spoken conversation through a dedicated ElevenLabs voice agent for each interview type, with real-time transcription and voice-activity detection handling the back-and-forth the way an actual phone or video interview would, not typed answers read back by text-to-speech.

The mechanics behind that matter more than they seem to. Detecting when a candidate has actually finished speaking, streaming audio without a lag that makes a conversation feel broken, transcribing accurately enough for grading to work from later, all of it happens underneath a single spoken exchange.

None of it is visible to a candidate mid-interview; what they experience is just a conversation that keeps pace with them.

What it is

Every one of Intervieux's seven interview types has its own dedicated ElevenLabs conversational voice agent behind it, distinct pipelines rather than one generic voice reading from a shared script.

Underneath that sit two more pieces doing real-time work during the conversation: Scribe for live transcription of what the candidate says, and Picovoice's Cobra for voice-activity detection, the system that recognizes when someone has stopped talking so the agent knows when to respond.

Audio itself streams over Socket.IO, and latency, how quickly the agent replies once a candidate finishes speaking, is treated as a first-class part of the engineering, not an afterthought bolted on later.

How it works

  1. 1

    Choose style and personality

    Three interview styles and twelve interviewer personalities set the tone the voice agent carries into the conversation before it starts.

  2. 2

    The dedicated voice agent opens the interview

    Whichever interview type was selected runs on its own ElevenLabs conversational agent built for that type specifically.

  3. 3

    Voice-activity detection tracks the exchange

    Cobra recognizes when the candidate has stopped speaking, which is what lets the agent respond at a natural point rather than talking over the candidate or leaving dead air.

  4. 4

    Audio streams in real time

    Socket.IO carries the audio exchange live, with latency engineered to keep the back-and-forth feeling like an actual conversation.

  5. 5

    Scribe transcribes as the conversation happens

    Real-time transcription writes down what was said as the interview goes, building the record that grading works from once it ends.

A dedicated agent per interview type

The voice agent for a technical interview isn't the same one running a salary negotiation session. Each of the seven interview types has its own dedicated ElevenLabs agent, which means the conversational behavior underneath the interview is built for that specific type rather than one voice reused everywhere with a different script pasted in.

Knowing when to respond

Voice-activity detection sounds like a small technical detail until it's missing. An agent that interrupts too early or waits too long after a candidate finishes talking breaks the feel of a real conversation immediately. Cobra handles that detection specifically so the timing of the exchange stays natural.

Latency as a first-class concern

How long it takes the agent to respond after a candidate stops speaking is treated as core engineering, not a detail to fix later. Audio streams over Socket.IO specifically to keep that gap short enough that the exchange still feels like talking to someone, not waiting on a system.

Where grading picks up from

The voice pipeline handles the live conversation; scoring happens afterward, on a transcript written by Scribe during the session and reviewed by a separate grading model with calibration built in. The two are connected but distinct: one runs the conversation, the other reads what came out of it.

Why voice instead of text

Typing an answer gives a candidate time to edit and reconsider before submitting. Speaking doesn't. A voice interview asks for the same kind of unscripted, real-time answer a candidate will actually have to give in front of a human interviewer, which is closer practice than composing a written response at their own pace.

Who it's for

Who it's for

This is the layer underneath every interview type on Intervieux, so it matters to any candidate practicing a spoken interview rather than typing answers into a form. It matters most to someone who's tried a lower-quality voice tool before and found the lag or the mistimed interruptions distracting enough to break the practice.

Frequently asked questions

Do all seven interview types use the same voice agent?

No. Each interview type runs on its own dedicated ElevenLabs conversational agent, built for that specific type rather than one shared voice.

How does the system know when I've finished speaking?

Voice-activity detection, running on Picovoice's Cobra, recognizes when a candidate has stopped talking, which is what lets the agent respond at a natural point in the conversation.

Is my interview transcribed as I speak, or afterward?

In real time. Scribe transcribes the conversation as it happens, and that transcript is what grading works from once the interview ends.

Does personality and style selection change the voice itself?

Style and personality shape the tone the voice agent carries through the same underlying pipeline, which is how the same interview type can feel formal and challenging in one session and friendly and encouraging in another.

Why does Intervieux use voice instead of typed answers?

Speaking an answer in real time, without the chance to edit before submitting, is closer to what a candidate actually faces in a human interview than composing a written response at their own pace.

Related pages

Hear the difference a real-time voice interview makes

Start any Intervieux interview type and talk through it live, with the same transcription and latency engineering behind every session.