Fish Audio Drama 3 Brings Multi-Speaker Voice Generation to SGLang

Published 17 October 2026

Fish Audio's 35B-parameter text-to-speech model supports up to 16 characters, voice cloning, natural-language controls, long-form streaming, and chained audio editing. Open-weight deployment is available through SGLang Omni, with single-stream first audio in under 100 ms on an H200.