Voice & turn-taking
Interrupt Finn.
Finn yields.
Natural turn-taking — when the caller speaks over Finn, Finn stops mid-word. No talking over each other. No awkward pauses.
30-minute free trial. No card. No procurement.
How it works
Listen even
while speaking.
Drag, wire, deploy. Each step is reversible — every save creates a versioned snapshot.
Bidirectional audio stream
Finn's mic stays open during synthesis. VAD detects caller voice the instant they start.
p50
384ms
p95
412ms
p99
538ms
Yield mid-word
Under 80ms from detected voice to TTS stop. No mid-sentence collisions.
p50
384ms
p95
412ms
p99
538ms
Resume context-aware
Acknowledges interruption, processes what the caller said, picks up the right thread. No 'as I was saying.'
Triage
Inbound router
DONEBooking
Scheduling specialist
ACTIVEConfirm
SMS follow-up
QUEUEDShared context · call IN-44219
caller.name = "Aarav Singh" caller.intent = "follow-up booking" captured.slot = "Tue 14:00" captured.doctor = "DR. Patel" captured.insurer = "BCBS"
Reads like real conversation
Caller leads.
Finn follows.
No bolt-ons. No second platform. Every primitive ships with the editor.
Sub-80ms yield
Faster than human reaction time. Finn drops the floor before the caller registers they're talking over.
Clean resume
Tracks what was said + what was cut off. Acknowledges + adapts. No robotic 'one moment please.'
Tunable barge-in threshold
Per-workflow VAD sensitivity. Quiet caller in a quiet room vs busy contact center — both work.
False-positive resistant
Trained to distinguish caller voice from ambient noise, music, cross-talk. Finn doesn't yield to a sneeze.
Shipped in production
Real teams, real outcomes.
“TOFA launched a national campaign with Finn — 10K+ calls daily and seamless human escalations.”
Alice Smith
Senior Engineer, Gofts · TOFA
Trust + ecosystem
Plugs in. Audits clean.
Wires into your existing stack via 100+ connectors. Compliant from day one across the workloads that demand it.
Twilio
PlivoElevenLabsOpenAI
Microsoft TeamsDon't see yours? Finn plugs into any system via REST + webhooks.
FAQ
Questions and answers.
What's the technical latency for barge-in?+
Sub-80ms from detected voice onset to TTS halt. Faster than the caller can perceive their own interruption.
How does Finn handle the interruption content?+
Captures what the caller said, processes it against the conversation context, and responds appropriately — apology, clarification, branch to new topic, whatever fits.
Can I tune sensitivity per workflow?+
Yes. Quiet inbound support calls use a low threshold. Noisy outbound to mobile users uses a higher threshold to ignore ambient noise.
What about cross-talk on bad connections?+
Echo cancellation + caller-voice classification filter cross-talk + line noise. Finn doesn't yield to its own voice bouncing back.
Does this work across all telephony providers?+
Yes — Twilio, Plivo, BYOC SIP trunk. Requires bidirectional streaming audio, which all major providers support.
Conversation, not monologue
Yields like a human.
Not a kiosk.
Real-time barge-in detection — Finn drops the floor the instant the caller speaks. No more 'press 1 to interrupt.'