Fish Audio Drama 3 Lands on AI Gateway: Direct Voice Performances with Natural Language

Fish Audio's text-to-speech model Drama 3 has officially launched on AI Gateway, kicking off a 30-day 80% discount. The model is currently in preview, but the capabilities it brings are already enough to turn heads in the audio creation world—you're no longer just reading text aloud, but directing the emotion, pacing, and performance style of every sentence through natural language instructions.

More Than Reading Aloud: Turning Scripts into Performances

Drama 3's core selling point is its "directability." Traditional TTS tools typically only care about whether pronunciation is accurate and whether the voice sounds pleasant, whereas Drama 3 lets you embed performance instructions directly into the script text.

The most basic usage is to mark emotion and pacing with square brackets, such as [warm and reassuring] Take your time. This wording makes the model deliver the line in a gentle, reassuring tone. You can also add sound effects and pauses in appropriate places to control the breathing rhythm of an entire speech segment.

Going further, you can wrap specific words with tags like <whisper> and <emphasis> to change how they're expressed. Put "I am right here" inside whisper tags, and the model will deliver it in a whisper—this kind of detail previously required a professional voice actor to adjust repeatedly to achieve.

TTS (Text-to-Speech) is the technology that automatically synthesizes input text into audible speech. It typically consists of two parts: a text front-end and an acoustic back-end. The front-end handles word segmentation, phonetic annotation, and prosody prediction, while the back-end generates a spectrum based on acoustic models, which is then converted into waveforms by a vocoder. Early TTS mostly used concatenation and parametric synthesis, resulting in mechanical voices and limited emotion. In recent years, end-to-end neural network models (such as Tacotron, VITS, and various LLM-driven speech synthesis approaches) have significantly improved naturalness. Traditional TTS control interfaces often only offer a few numerical parameters like speed, pitch, and volume, making it hard to express nuanced emotional layers. "Directable" models like Drama 3 bring performance instructions directly into the text input, making emotion, tone, and pauses writable, reproducible creative elements—essentially an expansion of the control dimensions of TTS.

Multilingual Instructions and Multi-Character Dialogue

Drama 3 has another easily overlooked but very practical feature: performance instructions can be written in a different language from the text being read. This means you can write directorial instructions in Chinese and have the model perform in English, or vice versa. For cross-national team collaboration or multilingual content production, this design saves a lot of communication overhead.

Multi-speaker dialogue is another key capability. The model supports assigning independent voices to each character, which greatly simplifies the production workflow for audiobooks, podcast dramas, game dialogue, and similar scenarios. In the past, you might have needed to generate audio for each character separately and then manually stitch them together; now you can complete multi-character dialogue in a single call.

Multi-Speaker TTS refers to using the same model to generate speech for multiple different voice timbres or characters. The traditional approach is to call the synthesis interface separately for each character, then use audio editing software in post-production to splice and align timelines—a cumbersome process prone to volume or noise floor inconsistencies at the seams. Multi-speaker models assign independent speaker embeddings or voice IDs to each character, completing an entire dialogue segment in a single generation, naturally ensuring voice distinction and temporal coherence between characters. For content formats with extensive dialogue—audiobooks, radio dramas, game dialogue, podcasts—this capability can significantly compress production cycles and makes it easier to uniformly adjust the overall pacing afterward.

How to Integrate: Code Examples and Model Name

Using Drama 3 on AI Gateway is very simple. The model name is fish-audio/drama-3, and the code structure is clear when calling through the AI SDK:

import { experimental_generateSpeech as generateSpeech } from 'ai';
import { writeFile } from 'node:fs/promises';

const result = await generateSpeech({
  model: 'fish-audio/drama-3',
  text: '[warm and reassuring] Take your time. <whisper>I am right here.</whisper>',
  voice: 'f3a2b90078d54a65af4b95f63bc7798e',
  outputFormat: 'mp3',
});

await writeFile('speech.mp3', result.audio.uint8Array);

As you can see, performance instructions are written directly in the text field, integrated with the reading content. The voice parameter specifies the timbre, and the output format currently supports mp3. If you don't want to write code, you can also try Drama 3 directly in the AI Gateway model playground.

What AI Gateway Provides

Drama 3's choice to launch on AI Gateway reflects the appeal of this gateway service for developers. AI Gateway provides a unified API to call different models, while also handling usage tracking, cost statistics, as well as retry, failover, and performance optimization configurations—with the goal of maintaining higher availability even when the provider itself experiences fluctuations.

It also includes built-in custom reporting, Zero Data Retention support, API key budget management, and routing rules. On pricing, AI Gateway charges at the provider's original price—no markup, no platform inference fees, including requests with bring-your-own-key (BYOK). For teams that need to control costs without sacrificing flexibility, these terms are quite friendly.

AI Gateway is a type of middleware service positioned between applications and model providers, with core value in "unified access + observability + governance." Developers only need to integrate one set of APIs to call different models from multiple vendors, without having to maintain separate keys, SDKs, and error handling logic for each provider. For stability, gateways typically include built-in retry, timeout, and failover mechanisms: when a provider experiences rate limiting or downtime, requests can automatically switch to a backup channel, reducing the impact of single points of failure on business operations. For cost and compliance, usage tracking, budget caps, and custom reports let teams clearly see consumption per key and per project; Zero Data Retention promises that request and response data are not persistently stored by the provider, which is especially important for scenarios involving sensitive information. Together, these capabilities turn "calling models" from a one-time coding task into long-term operational infrastructure.

Who Should Try It Now

Drama 3 is currently in preview, and the 80% discount will last for 30 days. If you work with audio content, character voiceovers, audio stories, or need to batch-generate emotional speech materials, now is a great time to validate at low cost whether it suits your workflow.

Its natural language directorial capabilities and multi-speaker support do push TTS one step further from a "reading tool" toward a "performance tool." Of course, the preview stage means stability and details may still have room for refinement—it's advisable to test the waters with non-critical projects first.

One-line Summary

Fish Audio's Drama 3 lands on AI Gateway, turning TTS into a directable performance tool with natural language instructions.

Paragraph Summary

Fish Audio's text-to-speech model Drama 3 has launched on AI Gateway with a 30-day 80% discount. Its biggest feature is pushing TTS from a "reading tool" toward a "performance tool": users can mark emotion and pacing with square brackets, change word expression with tags like whisper and emphasis, and even write performance instructions in a different language from the reading text. The model supports independent voiceovers for multiple speakers, simplifying production workflows for audiobooks, podcast dramas, and game dialogue. When calling through AI SDK, the model name is fish-audio/drama-3, and it can also be tried directly in the model playground. AI Gateway provides a unified API, usage tracking, failover, and Zero Data Retention, charging at the provider's original price with no markup.