Audio Signals
Examples: Barks, meows, whines, growls, pitch envelopes, duration, rhythm, and spectral shape.
Training use: Forms the core acoustic embedding used to predict likely intent and emotional state.
Collected Data
What we collect and why it matters
Context Inputs
Examples: Time of day, indoor/outdoor, activity mode, nearby stimuli, and optional owner notes.
Training use: Disambiguates similar sounds that mean different things in different environments.
Pet Profile Metadata
Examples: Species, age bracket, sex, optional breed, household structure, and recurring routines.
Training use: Improves personalization and helps tune model behavior across pet cohorts.
User Feedback Labels
Examples: Correct/incorrect confirmations, selected alternatives, and follow-up outcome notes.
Training use: Drives supervised fine-tuning and calibration adjustments for future versions.
System Telemetry
Examples: Inference latency, clipping/noise warnings, and failed or retried sessions.
Training use: Improves reliability, quality filtering, and infrastructure performance.