Xiaohongshu's FireRedTTS3 Unifies Speech Generation and Editing, Tops Benchmarks

机器之心 · wechat · 2026-08-21

The Xiaohongshu FireRed team released FireRedTTS3, a unified model for multilingual/dialectal cloning, natural language sound design, and precise speech editing.

Core Breakthrough: RedAE Representation

Architecture

Benchmark Results

Original post →

More from Multimodal

Multimodal channel →