Steps and facts
- The website voice demo demonstrates a discovery conversation: describe your business, try a customer scenario and see where AI and human handoff could help.
- In your workspace, Voice AI's Browser Simulator records the lead, transcript and handoff workflow. Use it to check knowledge and escalation behavior.
- Website Chat's Start talking option is browser speech input/output. It is different from the public voice demo and a telephone provider connection.
- WhatsApp calling, SIP and Ocktell are separate provider integrations and require configuration and end-to-end verification before being described as active.
- Test microphone access, the selected language, speech recognition, the spoken reply, stopping and interruptions on the browsers your customers actually use.
- Do not treat a successful text transcript as proof that live audio delivery or a telephone integration has passed.
- Facebook and Instagram voice messages: recognize one audio attachment up to 2 minutes and 8 MB and show its transcript in Inbox. The same approved knowledge, AI mode, confidence, human takeover and messaging-window rules apply. This supports recorded messages, not live Meta calls. Background processing can take about a minute; later text waits behind queued audio.
- Owners and admins can independently disable voice recognition or spoken replies under Configuration โ Channel connections. Both default to on and use workspace AI credit. Unclear or unsupported recordings go to a teammate. Eligible voice questions can receive a spoken reply; the text remains in Inbox. If generation fails before sending or Meta explicitly rejects audio, text can be sent instead. An uncertain delivery is not repeated as text. Test incoming audio and customer-side replies on the real account.