LLM to TTS
LLM to TTS Streaming
Pipe LLM token output directly to TTS as it arrives — the core latency pattern for voice agents.
Pipeline comparison
The table below compares a traditional full-text pipeline with the token-streaming approach used in real-time voice agents.
| METRIC | FULL-TEXT PIPELINE | TOKEN-STREAMING PIPELINE |
|---|---|---|
| Memory | Full response buffered | Constant |
| Complexity | Simple | Requires sentence buffering |
| Best for | Short known text | Voice agents, conversational AI |