Configuration for the expressive pipeline, passed as AgentSession({ expressive: ... }).

Controls how TTS markup instructions are injected into the LLM when expressive is enabled. All keys are optional; common shapes:

  • { speechSteering: {...} } — steer delivery and non-verbal sounds on top of the provider-agnostic default instructions.
  • { ttsInstructionsTemplate: '...' } — a fully custom prompt.
  • { ttsInstructionsAppend: '...' } — your own rules appended to the template.

Any explicit template overrides the default; unset parts fall back to the provider-agnostic default.

interface ExpressiveOptions {
    speechSteering?: SpeechSteeringOptions;
    ttsInstructionsAppend?: string;
    ttsInstructionsTemplate?: string | Instructions;
}

Properties

speechSteering?: SpeechSteeringOptions
ttsInstructionsAppend?: string
ttsInstructionsTemplate?: string | Instructions