Fish Audio recently raised a $52M seed round and reports 15M+ users. The model's integration with SGLang-Omni yields a median time to first audio of 89ms on a single H200 GPU. Performance increases to 72% higher throughput and 37% lower latency at 256 concurrent streams when moving the vocoder to a second H200 using SPG topology.
Tuesday, October 20, 2026
Story
Fish Audio Launches Drama 3 Open Weight Model for Word Level Control
AI7 days ago
Users can now dictate the precise emotional delivery of a single word in a voice clip and edit specific takes without regenerating the entire line. The Drama 3 text-to-speech model uses plain English instructions in square brackets to adjust tone, pacing, and style, allowing for attributes like whispers or forceful delivery. Fish Audio's model is open-weight and available through its app and API, supporting laughs, sighs, and multi-speaker dialogue.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.