vLLM团队分享了优化四种主流TTS模型服务的具体技巧，包括性能提升数据和实现细节，对部署TTS服务很有参考价值。
vLLM-Omni TTS团队针对Qwen3-TTS、VoxCPM2、Higgs Audio V3、Fish Speech S2 Pro四种TTS模型分别设计了不同的优化策略。对Qwen3-TTS通过解耦连接器分块和批处理Stage-0解码预处理，在H20×2上音频吞吐量提升61.5%，P99延迟减半。VoxCPM2采用whole-forward torch.compile和CFM/LocDiT解码尾部跨请求批处理，音频吞吐量提升172%。Higgs Audio V3将多码本解码状态机迁移到GPU驻留张量，实现2.7倍加速。Fish Speech S2 Pro为纯解码路径设计了模型特定的q_len=1 Triton注意力内核。
🎙️ Serving TTS isn't the same problem as serving a…
🎙️ Serving TTS isn't the same problem as serving an LLM. It has to hit a first-audio budget of a few hundred ms, keep audio continuous across streaming chunks, and sustain enough concurrent streams per GPU to keep serving cost down. It's also a multi-stage pipeline where each stage bottlenecks differently, so no single recipe carries across models. vLLM-Omni TTS team tuned a different lever for each of four TTS models:
🗣️ Qwen3-TTS: decouple connector chunking from the Code2Wav decode window, batch the Stage-0 decode preprocessing. +61.5% audio throughput on H20×2, P99 latency nearly halved. 🌊 VoxCPM2: whole-forward torch.compile, plus CFM/LocDiT decode-tail batching across requests. +172% audio throughput. 🎚️ Higgs Audio V3: move the multi-codebook decode state machine into GPU-resident tensors. 2.7x speedup. 🐟 Fish Speech S2 Pro: a model-specific q_len=1 Triton attention kernel for the pure-decode path.
Full engineering deep-dive on how we picked each lever: 🔗 https://t.co/ZVROwJwYoT