Create Expressive Voices with Drama 3

Fish Audio’s Drama 3 turns written text into spoken audio. It is now available through Vercel AI Gateway, a service for calling AI models. Drama 3 is still in preview, so it is not presented as a final release. The announcement offers 80% off for 30 days, through Nov 12.

The main feature is the ability to direct a voice in the script. Users can write instructions about emotion and speaking speed in square brackets. They can also add sounds and pauses where the script needs them. This means the input can describe both the words and how they should sound. The announcement suggests voiceovers, character conversations, and audio stories as possible uses.

Users can change how selected words are spoken with special tags. For example, <whisper> and <emphasis> tags direct a different kind of delivery. The directions do not have to use the same language as the spoken text. Drama 3 also supports conversations with more than one speaker. Each character can have a separate voice in the same conversation.

Developers use the model name fish-audio/drama-3 to access it through AI Gateway. An API lets one software service send requests to another service. AI Gateway offers a shared API for different models and helps developers keep track of usage and cost. It also provides request traces, which help show what happened during a request.

The announcement also describes budgets, routing rules, retries, and failover in AI Gateway. A retry sends a request again, while failover uses another option when a service fails. Gateway follows provider pricing without adding an inference markup or platform fee. The main change is that voice direction becomes part of the script. Developers can manage those model calls through the same gateway.