Apply queued commands and pending offset updates.

Process a single frame through the NLMS filter.

Process transcribed text through middleware pipeline. The pipeline now includes the provider middleware which calls Anthropic Claude

Run VAD on audio chunk. Args: chunk: Audio chunk (int16) Returns: True if voice activity detected,

Expose idle-related state for external threads.

Store a thread-safe snapshot of current frequency statistics.

Lightweight linear resampler for short buffers.

Root-mean-square in dBFS for float32 mono array in [-1,1].

Compute sleep time to maintain target frequency and update potential freq.

Set up PyAudio stream for microphone input. Returns: PyAudio stream object

Build the device auth block for the connect handshake. Returns the ``device`` dict to include in ConnectParams.

Convert arbitrary PCM array to float32 mono in [-1,1]. Accepts shapes: (N,), (1,N), (N,1), (C,N), (N,C).

Get face tracking offsets from camera worker thread. Behavior depends on head_tracking_mode: - "continuous": always read fresh offset

Update frequency statistics based on the current loop start time.

Advance queue state and idle behaviours for this tick.

Decode WAV and resample to target rate mono int16 PCM (best-effort).

Decode WAV and resample to 16kHz mono int16 PCM (best-effort).

Benchmark Kokoro TTS (requires espeak-ng).

Benchmark OpenAI TTS (requires OPENAI_API_KEY).

Benchmark Orpheus TTS (requires GPU + vLLM).

Abort an in-flight chat run. Best-effort. Returns True if the abort request was acknowledged.

Check required environment variables and dependencies. Raises: SystemExit: If critical requirements are missing

Clean up resources before shutdown.

Clear speaker buffer and reset filter (on playback cancel).

Close the WebSocket connection.

Combine primary and secondary full body poses. Args: primary_pose: (head_pose, antennas, body_yaw) - primary move secondary_pose:

Extract a speaker embedding from a WAV file for enrollment. Args: wav_path: Path to WAV file (any format — will be converted to 16kHz mon

Thread-safe: push base64-encoded audio into the consumer queue.

Thread-safe: extract PCM from WAV bytes and push into consumer queue.

Create a finish chunk.

Read and discard stale audio from the PyAudio input buffer. During TTS playback the VAD sleeps and doesn't read from the mic, so PyAu

Get the current date.

Get the current time.

Wait for VAD to detect speech (via queue). VAD runs continuously in background, this phase just waits for speech to appear i

Main entry point. Loads configuration, checks environment, and starts the state machine.

Extract (domain, dialog_name) from a SolveIt dialog URL.

Verify camera returns frames and SCRFD can detect a face. Returns True if a face was detected, False otherwise. Aborts the test early if the

Print startup banner.

Print usage instructions.

Process captured audio through STT pipeline. Flow: 1. Transcribe audio using local STT 2. Check for speech vs noise

Process user request through Anthropic Claude + middleware pipeline. VAD continues listening in background. If speech detected, inte

Reset the cancellation flag for reuse.

Reset accumulators while keeping the last potential frequency.

Reset the internal state.

Main loop: sleep, check presence, speak.

Register the CameraWorker instance for scene awareness tools.

Start the head wobbler loop in a thread.

Start continuous antenna wiggle animation (singleton). Waits for any in-progress reset to complete before starting. Saves cu

Initialize all components and prepare for operation. Transitions to INTERACTIVE state after completion (for debugging).

Signal the manager to stop.

Stop the head wobbler loop.

Test scene capture utility (requires robot camera).

Test SCRFD face detection on a synthetic image.