Add a `service` field so the frame targets a specific service, allowing
ServiceSwitcher.push_frame to consume it only when the targeted service
matches the active service. STTService and test mocks now push the frame
downstream after handling instead of silently consuming it.
Change the default user turn stop strategy from
TranscriptionUserTurnStopStrategy to TurnAnalyzerUserTurnStopStrategy
with LocalSmartTurnAnalyzerV3. Also reduce AUDIO_INPUT_TIMEOUT_SECS
from 1.0 to 0.5 and remove its debug log.
- Make ServiceSwitcherStrategy inherit from BaseObject with properties
for services and active_service, and move initial service selection
into the base class
- Add on_service_switched event to ServiceSwitcherStrategy
- handle_frame now returns the switched-to service (or None), allowing
ServiceSwitcher to swallow ManuallySwitchServiceFrame on switch and
request metadata from the new active service
- Override push_frame to suppress RequestMetadataFrame and
ServiceMetadataFrame from inactive services
- Remove ServiceSwitcherFilter and ServiceSwitcherFilterFrame in favor
of plain FunctionFilter instances with closures that check the
strategy's active service directly
- FunctionFilter: add FilterType alias
- FunctionFilter: when direction is None, frames in both directions
are filtered instead of just one
- Add docstrings to ServiceSwitcher and its components
Refactor TranscriptionUserTurnStopStrategy and TurnAnalyzerUserTurnStopStrategy
to use VADUserStoppedSpeakingFrame as the ground truth for when speech ended,
rather than triggering timeouts from transcription frames.
RTVIObserver now skips upstream frames to prevent duplicate RTVI messages
when frames are broadcast in both directions. Also changed
FunctionCallCancelFrame to use broadcast_frame for consistency with
other function call frames.
Processors inside parallel sub-pipelines can push frames during
StartFrame/EndFrame/CancelFrame processing. Previously these frames
could escape the ParallelPipeline before all branches finished
processing the lifecycle frame. Now they are buffered and flushed
after synchronization completes.
Move the interruption wait event from per-processor instance state to
the frame itself. The event is created in
push_interruption_task_frame_and_wait(), threaded through
InterruptionTaskFrame → InterruptionFrame, and set when the frame
reaches the pipeline sink. This scopes the event to each interruption
flow rather than sharing mutable state on the processor.
Also adds a 2s timeout warning to help diagnose cases where
InterruptionFrame.complete() is never called.
This adds user-to-bot response latency tracking to OpenTelemetry spans:
- Created UserBotLatencyObserver as a reusable component for tracking
user-to-bot response latency
- Records the value as an attribute on turn spans (turn.user_bot_latency_seconds)
- Updated TurnTraceObserver to use UserBotLatencyObserver, following the same pattern as TurnTrackingObserver
- Updated PipelineTask to automatically create and wire UserBotLatencyObserver
when tracing is enabled (same as TurnTrackingObserver)
