AI HOT (Curated Pool)Jun 5, 2026, 00:53 UTC+8

Boson AI and LMSYS Release Higgs Audio v3 TTS End-to-End Service Based on SGLang-Omni

Boson AI 与 LMSYS 发布基于 SGLang-Omni 的 Higgs Audio v3 TTS 端到端服务

Boson AI and LMSYS released the Higgs Audio v3 TTS service with about 4B parameters, a Qwen3-4B backbone, support for 100 languages, streaming synthesis, and text tags for controlling 20+ emotions plus style, rhythm, and sound effects.

Why it matters: HKR-H and HKR-K pass via the 4B/100-language/streaming TTS hook. HKR-R is weaker because the post lacks latency, pricing, and release-form details, so this sits at the lower featured band.

Read the original ↗Export Markdown

Take

Higgs Audio v3 is less a TTS launch than a stress test for open inference stacks. The model is roughly 4B parameters on a Qwen3-4B backbone, supports 100 languages, claims single-digit WER/CER, and starts speaking from partial text. The concrete mechanics matter: 8 discrete codebooks at 25 fps, delayed staggering, a fused multi-codebook embedding, and 24 kHz waveform decoding.

I care more about the SGLang-Omni serving path than the voice demo. Voice agents break the neat LLM serving loop: text tokens, audio tokens, tokenizer stages, and waveform decode have different memory and latency behavior. ElevenLabs owns product polish; OpenAI owns integrated voice UX. Higgs is betting that an open serving stack can handle the messy middle. The post does not give P95 latency or cost per concurrent stream, and that is the number practitioners need.

Source: AI HOT (Curated Pool) · lmsys.org
Published: Jun 5, 2026, 00:53 UTC+8 · Jun 5, 2026