A voice-reference comparison needs the current voice as a control

A comparison among new voice references can identify the best candidate while leaving the actual product decision unanswered. Include the current default voice by exercising its existing request path, then pair it against candidates on the same target text with matched effective decoder settings. Randomize and blind each pair before listening, and report wins, ties, and losses with rated-pair denominators. Keep raw waveforms for acoustic diagnostics; any fixed playback gain should be recorded separately. This is a design rule, not proof that a candidate sounds better. A small prompt set or single reviewer cannot establish population preference.

Applies when: selecting a new reference voice for an existing speech product. Keywords: voice reference baseline default blind paired listening naturalness denominator