pub struct CallConfig {
pub end_of_turn_ms: u32,
pub barge_in_ms: u32,
pub pause_ms: u32,
pub stt_mode: SttMode,
pub chunks: ChunkConfig,
pub partial_interval_ms: u32,
pub filler: String,
pub approval_reminder: String,
pub proposed_line: String,
pub failed_line: String,
pub warm_stt: bool,
}Expand description
Timings and fixed lines.
Fields§
§end_of_turn_ms: u32Silence after speech that ends the user’s turn.
barge_in_ms: u32Voice needed, over a reply, to stop it.
pause_ms: u32Silence that counts as a pause (and starts an early transcription,
or cuts a chunk). Keep it below end_of_turn_ms: a pause as long as
the end of the turn is never reported.
stt_mode: SttModeHow a turn is transcribed.
chunks: ChunkConfigWhere a streamed turn is cut (SttMode::Chunked).
partial_interval_ms: u32SttMode::Whole only: how often a turn is re-transcribed as a
caption; 0 is never. A chunked turn’s captions cost nothing extra.
filler: StringSaid during a tool round before any text; empty is nothing.
approval_reminder: StringSaid for an utterance heard while a card waits.
proposed_line: StringSaid when a turn stops at a card without having said anything.
failed_line: StringSaid when a turn fails without having said anything.
warm_stt: boolTranscribe a moment of silence as the call opens, and throw the words away. A self-hosted transcriber commonly unloads an idle model (the default for speaches is five minutes), and reloading it costs several seconds; warming it while the person is still drawing breath keeps that off the first turn.
Implementations§
Source§impl CallConfig
impl CallConfig
Sourcepub fn endpoint_config(&self) -> EndpointConfig
pub fn endpoint_config(&self) -> EndpointConfig
The turn detector timings these settings imply.