Do TTS wait for whole LLM response ?

Thanks for reaching out, Aman. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

yes share the options

You asked for the options to enable/tune streaming TTS with LLMs — here are the practical choices you can use to reduce latency and start audio as text arrives. • Plugin latency params ◦ ElevenLabs: set

streaming_latency

(seconds) — lower = faster first audio.

tts = elevenlabs.TTS(voice_id="ODq5z...", streaming_latency=1)

• Rime:

reduce_latency=True

disables text normalization for faster output (may affect pronunciation).

tts = rime.TTS(model="arcana", speaker="celeste", reduce_latency=True)

• Hume:

instant_mode=True

enables ultra-low latency (higher cost); auto-enabled when a voice is specified.

tts = hume.TTS(voice=hume.VoiceByName(name="Colton Rivers"), instant_mode=True)

• Choose streaming-capable TTS engines ◦ Use providers/plugins that stream audio (Cartesia, ElevenLabs, OpenAI, etc.) so audio is generated incrementally as tokens arrive. • Preemptive / incremental synthesis ◦ Let the TTS node synthesize chunks as the LLM yields tokens (TTS begins before the full response finishes). • Use LiveKit Inference / co-location ◦ Run models via LiveKit Inference for lower network latency and provisioned capacity (reduces TTFB and token latency). • Pick low-latency models ◦ Prefer models built for real-time (e.g., Cartesia Sonic, Rime Arcana) to reduce Time To First Byte (TTFB). • Monitor and tune metrics ◦ Track TTFB (first audio byte), TTFT (first token), and overall end-to-end latency. Adjust plugin params and model choices accordingly. • Avoid non-streaming models ◦ Don’t use TTS models that require full-text input before synthesis if you need low latency. Would you like example code for your stack (Node or Python) or help picking specific plugin settings for a particular provider? Sources: ElevenLabs TTS plugin guide | LiveKit Docs | Rime TTS plugin guide | LiveKit Docs | Hume TTS plugin guide | LiveKit Docs | Agents (Node.js) | livekit/livekit_composite | DeepWiki | Pipeline nodes and hooks | LiveKit docs