TTS streaming

Certain TTS providers, for example, ElevenLabs and Cartesia, support streaming mode. In this mode, TTS engine starts playing the LLM response before it’s completely generated, therefore, improving agent responsiveness and end-to-end latency.

You may use tts_stream advanced configuration parameter to activate the TTS streaming for your AI Agent.

Parameter	Type	Description
tts_stream	bool	If enabled, the LLM response will be streamed to the TTS engine.

For example:

{
    "tts_stream": true
}

Note that for TTS engines that don’t support streaming mode – e.g. Azure or Google Standard – this parameter has no positive effect. It may also interfere with other advanced configuration parameters that rely on complete LLM response being received before sending it to TTS, for example, llm_replace_words and llm_add_period, therefore, it’s kept off by default.