Xiaohongshu FireRedTTS3: Unified Speech Generation and Editing Model
jiqizhixin · x · 2026-08-26
Xiaohongshu's FireRedTeam presents FireRedTTS3, a unified model for cloning, designing, and editing speech across 24 languages and 21 Chinese dialects.
At its core is RedAE, a speech tokenizer that injects semantic meaning into acoustic representations from the start, eliminating the need for separate modules or multi-stage pipelines. The model handles zero-shot voice cloning from seconds of audio, text-described voice design, and precise speech editing where only the highlighted segment changes.
It achieved double first place in voice cloning accuracy and similarity (3.04% WER / 78.8%), double first in 24-language cloning (3.75% / 84.8%, top two in 22 of 24 languages), and leading scores on voice design and speech editing benchmarks.
More from Multimodal
- Open-source LoRA turns MiniMax H3 into seamless 360-degree orbit videos from a single photo — camenduru · 2026-10-15
- Neon Noir Elegance prompt template goes viral with striking pink-blue two-tone outputs — LudovicCreator · 2026-10-15
- Krea 2 Turbo + px16 LoRA generates game-ready pixel art in 5s, where GPT Image 2.5 fails — Angaisb_ · 2026-10-15
- How Goldfish Made Its Kung Fu Scenes: Real Nanquan Choreography + Seedance 2.5 — JunieLauX · 2026-10-15
- Custom px16 LoRA makes Krea 2 Turbo churn out pixel icons at 5s each — Angaisb_ · 2026-10-15
- Prompt share: 'Crimson Footsteps' makes one red element pop in monochrome AI photos — azed_ai · 2026-10-15