generated/voice_events.pb library

Classes

AssistantTokenEvent: Single token decoded by the LLM. is_final=true on the last token of a response (end-of-stream marker).
AudioFrameEvent: A chunk of synthesized PCM audio, ready for the sink. The frontend is expected to copy the bytes out; the C ABI does NOT retain ownership.
InterruptedEvent: Assistant playback was interrupted by a barge-in. The reason distinguishes user barge-in from app-initiated cancel.
InterruptReason
MetricsEvent: Per-primitive latency breakdown. Emitted at barge-in and at pipeline stop.
PipelineState
StateChangeEvent: Pipeline lifecycle state. Ordered — callers can compare numerically.
TokenKind
TurnLifecycleEvent
TurnLifecycleEventKind
UserSaidEvent: User speech finalized by STT (is_final=false → partial hypothesis).
VADEvent: Voice Activity Detection output. Frontends usually do not need this — exposed for debugging and custom UIs (waveform highlighting, etc.). type uses the canonical VADStreamEventKind enum from vad_options.proto (the hand-rolled VADEventType was deleted).
VoiceAgentComponentStates: Aggregate load state across all four voice-agent components. Mirrors Swift VoiceAgentComponentStates, Kotlin VoiceAgentComponentStates, RN VoiceAgentComponentStates, Web VoiceAgentComponentStates, and Flutter VoiceAgentComponentStates.
VoiceEvent
VoicePipelineComponent
VoiceSessionError