Voice messages

Customers can send voice messages, and agents can reply by voice. The two sides don't need to speak the same language: the customer speaks English, the agent speaks Chinese, and the customer still hears English.

A customer sends a voice messageCopy Anchor Link

Voice messages come in from:

• The website chat window: the customer taps the microphone next to the message box and records for up to 60 seconds.
• WhatsApp, Instagram and Messenger: the customer's voice messages.

purser turns the voice message into text, and from then on it is handled like typed text:

• The AI answers the words, rules check them, and they are translated into the agent's work language as usual.
• The inbox shows the recording, ready to play, with the transcript and its translation under it.
• In the chat window, the customer sees their own recording with a line "Heard: …" under it.
• If the conversation's language isn't known yet, it is taken from the language heard in the recording.
• A voice message recorded in desktop Chrome is WebM, which may not play on an iPhone (Safari and the agent app); the transcript is shown as usual.

If nothing could be made out (silence or noise), the AI doesn't answer. The conversation goes to a person, and the inbox says "The customer sent a voice message we couldn't make out. Please listen to it."

If the channel couldn't hand the recording over, the conversation gets a note asking you to have the customer send it again or type.

Agents reply by voiceCopy Anchor Link

In the inbox, click Speak next to the reply box, say your reply in your work language, and click again to stop. Each recording can be up to 60 seconds.

1. purser turns what you said into text and translates it into the customer's language. The reply box gets the translation, with your own words under it.
2. Read it over and edit it if needed.
3. A button such as Send as voice (English) is switched on for you: when you send, the customer first gets a voice message, then the same text. Click the button again if you don't want to send the voice message.

Your recording is only used to get the words. It isn't kept and isn't sent to the customer. The customer hears speech that purser makes in their language.

You don't have to record: type your reply and switch on Send as voice, and what you typed goes out spoken.

Which languages and channels can get voice repliesCopy Anchor Link

Email replies are never spoken. Where a voice reply isn't possible, the Send as voice button isn't shown. If the voice can't be made, the reply goes out as text only and the conversation gets a note.

Transcription (turning speech into text) covers far more languages: whatever language the customer speaks, and whatever work language the agent speaks in.

The AI's own replies are still sent as text only.

How long recordings are keptCopy Anchor Link

• Recordings are kept for 90 days, then deleted. The inbox then shows "Recording expired"; the transcript stays.
• Recordings can only be played through links that last 10 minutes. In the chat window, a customer can only listen to recordings in their own conversation.
• Voice replies sent to WhatsApp, Instagram and Messenger use a link that lasts one hour, for Meta to fetch.

In an appCopy Anchor Link

The microphone on the app chat page (/c/…) needs the app to let the web page use the microphone; see In-app support page.

The chat window on a store's pages asks for the microphone on its own. If the customer says no, or the store's site doesn't allow the microphone in embedded pages, the window asks the customer to type instead.