Research | Model Progress

Current Model Status And What Is Improving Next.

PetSpeak is in active development. We publish progress so users can understand where interpretation quality is strongest today and where it remains experimental.

Back to Research Subscribe to Updates

Benchmarks

Performance snapshot

Interpretation Match Rate

68-74%

Range across controlled test scenarios with user-validated context.

Calibration Stability

0.12 ECE

Confidence scores are still being tuned for calmer/low-intensity sessions.

Median Latency

1.4s

Web translation turnaround from recording submission to output.

High-Noise Degradation

-15%

Typical drop in match quality when audio includes overlapping voices.

Dog Coverage

Beta

Largest current training set; broader behavior class separation in progress.

Cat Coverage

Early Beta

Strong on repetitive domestic contexts, weaker in unfamiliar environments.

Roadmap

Development focus by phase

Phase 1

Q2 2026

Stabilize core signal decoding for recurring home contexts and routine triggers.

Phase 2

Q3 2026

Improve confidence calibration and expand breed/context representation in training sets.

Phase 3

Q4 2026

Ship stronger multi-pet environment handling and team analytics for pro users.

Interpretation Limits

How to read outputs responsibly

Probabilistic output

Translations are estimated meaning bands, not literal language conversion.

Confidence context

Higher confidence is more reliable when clip quality and context inputs are both strong.

Safety behavior

Aggression or distress indicators should always be treated as cautionary and observed directly.