SpeechSettingsInput
Speech output settings.
An object with 5 fields.
Fields
closing_linestringSpoken before the Call ends at its duration ceiling (
max_duration).greetingGreetingInputlimit_messagestringThe line spoken when a limit stops a reply (1.24.0): after a Policy's
block or end_call without its own say, and when the model reaches the Tool round cap (max_tool_rounds), through before_speak with the source fallback (agent-controls.md section 4). A template of at most 300 characters in the Agent's language. An Agent version may refer to its declared variables as {{variables.<name>}}, as in the greeting text, and the api renders them at admission; the Organization layer may not refer to a variable (variable_unknown). Omitted at every layer, the language pack's line, which says the Agent cannot help further with that and offers to help with something else. Policies and Tools are Phase 2, so in Phase 1 the line is stored, validated and resolved, and no Session speaks it. It is not the Fallback line, the human line of a Phone number.recovery_linestringThe line spoken after a model error in the middle of a reply (1.24.0, voice-plane.md section 7.3). A reply the caller has partly heard is never retried on another Provider: it ends at what was spoken, the Session speaks this line through
before_speak with the source recovery, and the next Turn uses the next healthy Provider. The line is also spoken when a model request fails before any word and every Provider of the llm chain has failed. A template of at most 300 characters in the Agent's language. An Agent version may refer to its declared variables as {{variables.<name>}}, as in the greeting text, and the api renders them at admission; the Organization layer may not refer to a variable (variable_unknown). Omitted at every layer, the language pack's line, which apologises for a brief problem and asks the caller to say it again.speaking_ratenumberA multiplier of the voice's normal rate, where the voice Provider declares a rate control; ignored and recorded otherwise.