Direct delivery in ordinary language

Fish Audio is previewing Drama 3, a text-to-speech model that accepts performance direction in ordinary language. Describe a character’s tone and pace instead of assembling audio tags. The company also says delivery can change within a sentence, so a line can begin calmly and end with urgency.

Drama 3 appeared earlier in a creator-only soft launch. The newer preview announcement adds a developer model name and describes more specific editing controls.

Several characters, one scene

Fish Audio says Drama 3 can generate multi-character scenes. One possible use is an animation scratch track: write a short exchange, give each speaker a different attitude, then listen for the rhythm before recording the final cast. This is a suggested use, not a XANTRANS test.

For example, a direction such as “The shopkeeper stays matter-of-fact; the customer grows increasingly impatient” gives the scene a performance goal. Whether the voices maintain their identity across a longer project still needs testing.

Repair a word instead of restarting

The preview also promises single-word fixes or regeneration. That could matter when a name or product term changes after the rest of a read has been approved. Instead of choosing a whole new take, an editor could try replacing only the problem word.

The announcement does not explain whether nearby breaths or speech can also change. Listen across the edit boundary before treating a retake as finished. The useful result is a correction that preserves the surrounding performance.

Preview access and open questions

The vendor announcement names drama-3-preview for API requests. Check preview access with Fish Audio before planning an integration. Its public models overview still recommends S2.1 Pro for production and does not list a comparable Drama 3 specification.

Drama 3-specific pricing, free-plan entitlement and language coverage are not established by the linked announcements. Treat it as a preview for evaluation, rather than assuming the terms of another Fish model apply.

Sources & verification

Written with AI assistance from Fish Audio's public announcements and documentation. XANTRANS has not independently tested Drama 3.