Fish Audio Releases Drama 3 Open-Weight Emotive TTS Model and Announces $52M Seed Round

On October 14, 2026, voice AI startup Fish Audio officially introduced Drama 3, an open-weight text-to-speech (TTS) foundation model engineered for fine-grained word-level emotion control and dynamic mid-sentence vocal shifts. Alongside the model release, the company announced surpassing 15 million users and securing a $52 million seed funding round.

Image source: Fish Audio (@FishAudio) on X

While conventional speech synthesis systems have largely restricted emotion control to broad sentence-level styles or fixed speaker embeddings, Drama 3 introduces granular acoustic modulation—allowing creators to dictate how individual words resonate and transition vocal delivery within a single spoken line.

Word-Level Modulation and Mid-Sentence Voice Transitions

Positioned by Fish Audio as "the most emotive and controllable TTS model in the world," Drama 3 centers its technical appeal on granular acoustic controllability.

• Word-Level Pronunciation and Pitch Control: Creators can isolate single words within a passage to modulate inflection, raise pitch, or drop into a whisper. This capability targets dubbing, game character dialogue, and podcast narration where precise contextual pacing is critical.
• Mid-Sentence Voice Shifts: The model allows shifting vocal timbre seamlessly within a single uninterrupted sentence. This enables dramatic expressions with abrupt emotional pivots, internal monologues, or multi-character voice transitions without splicing separate takes.
• Open-Weight Architecture: Rather than locking developers into a proprietary hosted API, Fish Audio released Drama 3 with an open-weight architecture, allowing organizations and researchers to inspect, fine-tune, and deploy the model within private infrastructure.

Alongside the model rollout and an official 4K showcase video, Fish Audio highlighted that creator Uncle Roger "was impressed too" with the new model.

Day-1 Integration Across 14 Platforms and SGLang Inference Support

Alongside the model debut, Fish Audio announced an extensive launch ecosystem spanning creative tools, API routers, and serving engines.

A total of 14 partner platforms provide Day 1 integration:

• Creative and Generation Platforms: ComfyUI, Runway, Pika Labs, OpenArt, Creatify, TapNow, Astorie, and OpenClaw
• Model Routing and Cloud Infrastructure: OpenRouter, Vercel, OrcaRouter, Runware, Inference.ai, and NousResearch (Hermes agent)

In parallel, inference frameworks SGLang (@sgl_project) and RadixArk (@radixark) rolled out Day-0 support for Drama 3's open weights. This enables developers and enterprise teams to deploy scalable speech pipelines on open-source serving infrastructure immediately upon release.

$52M Seed Funding and Production Deployment Considerations

The $52 million seed round reinforces Fish Audio's ambition to expand its voice foundation model stack across creative and commercial workflows.

For engineering teams evaluating Drama 3 for production pipelines, several verified operational factors warrant review:

• Licensing and Weights Distribution: Fish Audio paired the release with a promotional campaign (offering 1 month free for users commenting "DRAMA" within 48 hours and a complimentary first month for enterprise accounts). Specific model repository locations (such as Hugging Face) and precise commercial usage license terms should be verified as release documentation finalizes.
• Serving Infrastructure and VRAM Budgets: Delivering real-time inference with fine-grained word-level modulation requires dedicated compute. Recommended GPU memory specifications and throughput characteristics will vary depending on the serving engine (SGLang, RadixArk) and applied quantization schemes.

Sources

• Fish Audio Official X (@FishAudio): Drama 3 Model Announcement and $52M Seed Round Post
• Fish Audio Official X (@FishAudio): Drama 3 Day-1 Launch Partners Ecosystem Announcement
• Fish Audio Official X (@FishAudio): Drama 3 Day-0 Inference Support on SGLang and RadixArk