verify

To verify Rust port accuracy, you compare generated audio and model latents against a known-good Python version using correlation metrics. This Skill automates that numerical accuracy testing to ensure the port matches the original model's behavior.

Audio correlation testing measures the numerical similarity between audio outputs of two model implementations. You need it for ML model ports to detect regressions and confirm that the ported version produces audio with at least 0.95 correlation to the reference.

You benchmark TTS synthesis latency by measuring Time To First Audio (TTFA) and Real Time Factor (RTF) during the text-to-speech generation process. This Skill runs those performance benchmarks automatically alongside accuracy checks.

Yes, the verification process relies on Whisper and wavfile dependencies, alongside numpy, scipy, and wavfile, to load audio, compute correlations, and assess the numerical accuracy of the Rust port against the Python reference outputs.

A Rust TTS port might produce different audio due to numerical regressions or code changes. Running this Skill's regression detection identifies degradation by measuring audio and latent correlation against the Python reference to pinpoint discrepancies.

Yes, you can use this Skill for continuous integration and regression detection in ML model ports. It assesses build status, measures synthesis latency, and validates audio correlation to catch any degradation after code changes automatically.