Problem it solves
Text inboxes are saturated; voice messages stand out because almost no one sends them, but recording one by hand per prospect, with retakes, is impractical. The voice step lets the rep generate a voice note from written text using a library voice, so they get the cut-through of voice without recording anything, or record their own audio if they prefer.
Use cases supported
Adding a voice message step to a campaign sequence without recording audio manually (Generate with AI).
Recording your own voice note for the step instead, if you'd rather (Record Manually).
Picking a voice from the library and tuning how it sounds (speed, voice stability, background noise) before previewing.
Writing the message text with personalization attributes, previewing the result, then keeping the step in the sequence.
How it works
Where to find it. It's a campaign step, not a standalone page. In a campaign, open the Editor, add a step, and under Actions choose "Send LinkedIn Voice Message." (In the step list it sits near Send LinkedIn Connection / Send LinkedIn Message / Send LinkedIn InMail / Send Email.) The voice step is typically placed after an "Accepted LinkedIn connection" condition, so it only fires for prospects who've connected.
The step panel. Selecting the step opens a panel with two tabs:
Generate with AI (the AI path). Controls, in order: Select a voice (searchable dropdown of library voices); a Background noise on/off toggle ("Make your messages feel more natural"); a Speed slider (0.5x to 2x, default x1); a Voice Stability slider (Variable to Robust); and a Message field where the rep writes the text, with Add Attribute for personalization and a Preview to hear the result.
Record Manually (the alternative path). The rep records their own audio for the step instead of generating it.
The rep writes the text, picks a voice, tunes the controls, previews, and keeps the step in the sequence. A freshly added step shows an "Action required" state until it's configured (e.g. a voice and message are set).
Known limitations
Two modes: Generate with AI uses library voices (not voice cloning); Record Manually uses the rep's own recorded audio. Be precise: the AI path does not clone or synthesize the rep's own voice, it uses a selectable library voice. If the rep wants their own voice, that's the separate Record Manually tab (a real recording, not AI). Do not blur the two or imply the AI generates "your voice."
It's a campaign step, configured per sequence. There's no global "AI Voice" page; the rep sets it up inside the campaign each time. Don't describe a standalone feature surface.