Answer
You can ask for a voice note, request “Read to me from this text,” or ask AI to create and perform a two-person dialogue. To hear an earlier answer, reply to it and ask “Read this aloud.” The written response appears first, and the MP3 follows in the same answer when it is ready; press Play then. Requests such as “Create a voiceover for this paragraph” and “Give me an audio version of this answer” also request speech.
Text can stay selected. Audio also supports spoken answers: open the model dropdown beside @ai, choose Audio Generation → Music and make the same explicit speech request. This works in private and group chats on web, phone browsers, iPhone and Android, without a voice picker or live call.
Your request guides language, tone and pace. Limits are 450 words, 6,000 transcript characters and three minutes of audio. A request outside the supported limits may receive a written explanation. Ordinary Markdown can use one narrator. Speech does not clone a voice or create music. Asking a question about speech, quoting instructions or requesting only a script does not request audio.