assets

var AudioForcedStopULaw []byte

AudioForcedStopULaw is the forced-stop clip (SOP-156), μ-law encoded at 8 kHz (1 byte/sample). It plays when a single utterance exceeds MaxUtteranceMS and the caller is cut off, then the call terminates through the same mark-echo flow as the farewell. Like FarewellULaw, nothing depends on its length: MarkEchoTimeout derives the post-playout wait from it.

var FarewellULaw []byte

FarewellULaw is the call-termination farewell clip (SOP-125), μ-law encoded at 8 kHz (1 byte/sample, so len/8000 == seconds). It says "Call me anytime... Bye!" in the voice-out TTS's F5 voice.

voice-out is the engine that will speak the engine's replies once SOP-115/G,H builds that path, so sourcing the farewell from it means she will end a call in the voice she spoke it in. Today it is the only thing she says: handleSTTResult logs a transcript and returns, and nothing in this package talks to voice-out at all.

Regenerate with scripts/make_farewell.sh (see PROVENANCE.md). Nothing depends on its length: MarkEchoTimeout derives the post-playout wait from len(FarewellULaw), so a longer or shorter clip retimes itself.

var LLMThinkingULaw []byte

LLMThinkingULaw is a loopable "thinking" bed, μ-law at 8 kHz. SOP-156 embedded it with no playback path yet; SOP-157 plays it on the outbound data plane during the sim-turn collection stand-in (Session.sendBed), looping to fill the configured duration ahead of the real LLM-composed reply (SOP-115/G,H), which will eventually replace this fixed clip.