prelulu LISTENING ROOM VOICE STUDY 01
S2 PRO & S2.1 PRO · BLIND COMPARISON

Familiar characters.
A closer listen.

Forty popular characters. Two speech models. The same voice and words in each pair. Listen first, then reveal what Gemini heard.

20 male characters20 female charactersGemini 3.1 Pro judge
—CHARACTERS
—MATCHED PAIRS
—BLIND REVIEWS

Matched inputs.
Reversed listening order.

01 /

Make up your own mind.

Model names and judge scores start hidden. Your picks stay in this browser.

02 /

A fair pair.
An honest result.

01

Choose the voices

Twenty popular male characters and twenty popular female characters, selected using recorded usage.

02

Hold the inputs still

Each pair uses the same character reference voice, text, and synthesis settings. Only the requested model changes. Audio is generated specifically for this study.

03

Listen without labels

Gemini 3.1 Pro hears anonymous A/B samples. Each comparison is repeated in reverse order to expose order-sensitive judgments.

04

Keep the limits visible

One automated judge is not a human preference study. A controlled sample does not prove live-call latency, reliability, retention, or conversion.

Study notes & reproducibility
Study report Results CSV Protocol Gallery JSON