Fish Audio has opened a preview of Drama 3, a text-to-speech model whose pitch is that you direct a line the way you would direct an actor. Direction lives in the script as plain text. Where a sound or mood should land, the user types it in square brackets, for example [sigh]. Longer, free-language descriptions of the delivery also work, written in English or another language. Paired tags like Whisper or Slower limit an effect to selected words. An Auto Tag button drafts directions, and a multi-speaker mode performs a whole dialogue at once.
It is live in the company’s Text to Speech editor. Developers can reach it through the API by naming drama-3 in the request header. The page gives no pricing, latency, language count or cloning details, and says nothing on consent or safeguards, though its demo uses a licensed Uncle Roger voice. Fish Audio concedes results vary between takes. Every claim comes from the company’s own page; no independent evaluation is cited.
The test for anyone producing audio is how many takes it takes to land a single direction.
Based on Fish Audio’s Drama 3 product page, undated, retrieved 14 October 2026
