Test face tracking with real robot camera.

Convert to conversation message format. Returns: Message dict with role and content

Animation based on tool type. Tool animations disabled - will be re-enabled later.

Create a tool notification chunk (for animations).

Process audio sample and return VAD confidence. Args: audio_sample: Audio with shape (batch_size, num_samples)

Clean up ONNX session.

Initialize the audio input processor. Args: stt_engine: Local STT engine instance (required)

Args: frame_size: Samples per AEC frame (160 = 10ms at 16kHz). filter_length: NLMS filter taps (3200 = 200ms echo tai

Args: queue_move: Callable that queues a Move on the MovementManager. Signature: queue_move(move: Move) -

Initialize a stream chunk. Args: content: Text content of the chunk role: Role of the message sender (e.g.,

Initialize the state machine with configuration. Args: config: Configuration dictionary from config.yaml

Args: send_message: Async callable that sends a message through the chat pipeline. Signature: async def

Initialize VAD with ONNX model. Args: model_path: Path to silero_vad_v5.onnx model

Initialize VAD speech capture. Args: audio_stream: PyAudio stream for reading audio sample_rate: Audio sampl

Initialize filter with a predicate function. Args: predicate: Function that returns True if chunk should be yielded,

Initialize with sentence-ending punctuation set.

Initialize TTS middleware. Args: tts_url: URL of the TTS service endpoint timeout: Request timeout in second

Initialize interruptible TTS middleware. Args: tts_url: URL of the TTS service endpoint timeout: Request tim

Initialize robot animator. Args: mini: ReachyMini instance, or None for mock mode

Initialize state.

Initialize breathing move. Args: interpolation_start_pose: 4x4 matrix of current head pose to interpolate from interp

Initialize movement manager. Args: head_tracking_mode: "continuous" (always follow face), "orient_on_speech" (sna

Initialize the head wobbler.

String representation for debugging.

Blocking antenna wiggle implementation. Runs in thread executor to avoid blocking async loop. Args: duration: D

Call TTS service via HTTP asynchronously and play the audio. Args: text: Text to synthesize and play

Handle SIGINT gracefully — stop the run loop so cleanup runs.

Convert float64 array [-32768, 32767] to int16 PCM bytes.

Speech onset callback — fires when VAD first detects voice. In INTERACTIVE: orient head toward speaker (at s

Send a spontaneous message through the chat pipeline (used by SpontaneousSpeechManager).

Generate and speak a sarcastic GLaDOS startup announcement. Uses the LLM to create a dynamic response each time.

Blocking wake word detection (runs in thread executor). Returns: True if wake word detected, False if interrupted

Add a middleware to the pipeline.

Quick antenna wiggle animation (for interruption). Args: duration: Duration per antenna movement in seconds

* Create and configure an APM instance. * * sample_rate: 8000, 16000, 32000, or 48000 * num_channels: typically 1 (mono) * * Returns an opaque

* Destroy the APM instance and free resources. */

* Process near-end (mic/capture) audio through the APM, in-place. * * handle: opaque handle from apm_create() * near_end_data: pointer

* Feed far-end (speaker/render) audio into the APM. * * This must be called BEFORE apm_process() for each time-aligned frame. * The APM uses this t

* Reinitialize the APM (reset internal state without destroying it). * Useful for clear() operations. * * Returns 0 on success. */

Get tool arguments.

Get audio data.

Stop scheduling further sentences and clear the audio buffer.

Cancel current playback and clear queue (for interruption). Waits for playback to actually stop before returning.

Stop scheduling any further sentences. Note: cannot stop an already-playing sentence WAV (testbench has no stop API).

Build a message content array with the current camera frame + question. Returns a list of content blocks suitable for Anthropic's messages API:

Clear buffered speaker reference and reset the canceller state.

Get chunk content.

Create a RobotAnimator instance. Args: mini: ReachyMini instance, or None for mock mode Returns: RobotAnimator instance

Duration property required by official Move interface.