Validate speech quality scores against human phone controls

A speech quality model trained outside a narrowband call path may rank generated speech above known human recordings. Before using its score to choose a voice or decoder setting, process human and synthetic controls through the same resampling and codec path and check that it recovers the broad expected ordering. If it fails, keep pitch, ASR, speaker similarity, and waveform measures as anomaly detectors rather than claims of naturalness. Different sentences and recording conditions make this a sanity check; a calibrated selector still needs human ratings on matched phone audio.

Applies when: optimizing synthesized voice quality for 8 kHz telephone delivery. Keywords: TTS naturalness MOS narrowband phone codec human control scorer validation