Voice
Let customers talk to the Informly AI assistant through their browser with no phone number, no SIP, and a choice of natural-sounding voices.
Voice lets your customers actually talk to the assistant — they click a button in the widget, grant microphone access, and have a spoken conversation that the AI handles in real time. There's no phone number to provision and no SIP trunk to wire up; the whole thing runs in the browser over WebRTC.
Behind the scenes, Informly uses LiveKit for the audio transport, with speech-to-text and text-to-speech handled in the cloud. Customers hear the assistant in a voice you pick from the Voices library in the sidebar (Google's voice catalog powers the TTS). A human agent can join the call if the AI gets stuck — see Voice calls in handoff.
Where to find it
Go to Widgets → (your widget) → Voice.
Picking voices
Browse the voice library
Open Voices in the sidebar to listen to samples. Each voice has a language, accent, gender, and short preview clip so you can hear how it sounds before assigning it.
Assign one or more voices
Back on the widget's Voice tab, tick the voices you want available. Most teams assign a single voice that matches the persona, but you can assign multiple if you want the customer to choose.
Save
Click Save. Voice becomes available to that widget's customers immediately.
What the customer sees
When voice is enabled, a microphone or "Call" button appears in the widget. Clicking it:
- Prompts the browser for microphone permission
- Opens a voice session over WebRTC
- Starts streaming the customer's audio to the speech-to-text service
- Sends transcribed messages to the AI, which replies in text
- Synthesizes the AI reply with the voice you picked and plays it back
The customer can switch back to text at any point in the same conversation; the transcript carries over.
Cost
Voice synthesis is billed per character of generated speech. A short three-sentence reply might cost less than a cent, while a long monologue costs more. Two things keep cost in check:
- Keep persona prompts tight so the AI doesn't ramble
- Watch the analytics for any widget that's producing unusually long replies
Use the voice synthesis estimate tool in the cost dashboard to forecast spend before turning voice on for a high-traffic widget. See Analytics → Cost for the running total once you're live.
Voice spend is one of the easiest line items to underestimate. A single popular widget can generate millions of synthesized characters a month. Set a budget alert in Billing before enabling voice on a high-traffic page.
Browser support
Voice works in any modern desktop or mobile browser with WebRTC support. The customer needs to grant microphone permission; the prompt appears the first time they click the call button. If the customer denies permission, the widget gracefully falls back to text.
Human handoff during a call
If the AI hits a question it can't answer or the customer asks for a person, a human agent can join the live audio call from the Informly inbox. The AI mutes itself, the agent takes over the conversation, and the transcript continues to be recorded.
See Handoff → Voice calls for the agent-side experience and routing rules.
What's next
Persona
Make sure the assistant's written tone matches the spoken voice you picked.
Voice calls in handoff
Set up rules for when a human agent joins a live voice call.
Cost analytics
Track per-widget voice spend before it surprises you.
Security
Restrict voice to specific origins and rate-limit abusive sessions.