Voice messages

Customers on Messenger, WhatsApp, Telegram and Viber often send a voice note instead of typing. The agent can listen and answer the same way.

What happens

1. The voice note is transcribed to text.
2. The transcript is handled exactly like a typed message — knowledge, tools, routing, handover.
3. The language is detected from the transcript.
4. The reply is spoken back in that same language, with the text included too.

A customer who sends a voice note in Burmese gets a spoken reply in Burmese. Someone who types gets a typed reply — the agent answers in the form the customer chose.

Turning it on

Both directions need nothing extra — understanding a voice note and speaking a reply back both run on the same AI that answers your messages, so they work as soon as the agent does.

When it is on, a customer who sends a voice note gets both: the reply written out, and the same words spoken. The text is never replaced — it has to stay readable for the customer in a quiet room and for whoever reads the case back later.

The spoken voice matches the gender you set for your agent, because Burmese and several other languages mark the speaker's gender in ordinary words. A language with no voice available is answered in text rather than in a voice that cannot pronounce it.

A voice note is never guessed at. If it cannot be understood — a bad recording, or a language the model cannot make out — the case goes to a person and says why, rather than the agent answering something the customer did not say.

The Media Library page shows the current state at a glance:

Transcripts

Every transcript is stored on the message, so the inbox, search and reports all read voice conversations as ordinary text.