How to make a bot reply with voice messages

The assistant takes the text it was about to send, reads it aloud and drops it into the chat as a voice message. To the other person it is an ordinary voice note — they listen to it rather than read it.

Where to turn it on

The first screen of the app has a “How to reply” block with three buttons.

• “With text” — the assistant understands incoming voice notes but answers in writing.
• “Match the sender” — voice for voice, text for text. This is the default.
• “With voice” — every reply is spoken aloud, including replies to plain text.

The choice saves immediately. Nothing needs restarting — the next incoming message is handled the new way.

Which voice

The voices live in the “Voice replies” section. The list is live — it comes from the synthesis service — and every voice has a play button. Spend a minute listening: you hear exactly what your client will hear.

Next to it are speed — from 0.5 to 2 — and pitch and volume, both from −10 to +10. You do not have to touch them; the defaults work.

What one reply costs

Synthesis is billed by character count, so the price depends on both the voice and the length of the reply. Under each voice in the list it says how many Stars one reply will cost at your current length limit. The spread is wide: between the cheapest and the most expensive voice the difference is more than twentyfold.

The limit itself is a slider from 100 to 900 characters in steps of 50. It trims both the text and the audio, which makes it the only setting that controls the price of a reply directly.

What happens to a long voice note

The assistant does not transcribe an incoming note longer than a minute. The sender gets this reply:

That voice message came out longer than a minute — I cannot handle those yet. Could you send a shorter one, or write it out?

If the recording is silence or solid noise, the assistant will say plainly that it could not make it out and ask for a repeat. It will not guess the content from the first few seconds.

If the audio fails

The reply goes out as text. Voice is built as a layer on top of ordinary sending: a synthesis failure does not eat the message, and the client still gets an answer — just in letters. If the synthesis keys are not connected, the section is switched off entirely and the assistant works as before.