Two numbers instead of one is the part most voice comparisons miss. A voice that starts fast but finishes slowly feels fine in a chat and drags in a long narration. I'd add two things to the review step. First, match loudness before comparing takes, because the louder of two takes almost always sounds better on a quick listen, whatever the voice. Second, hide the voice name while listening and only reveal it after you've rated the row. Otherwise it's easy to end up confirming the voice you already liked on the provider's demo page. The getAll check for exactly one voicePreset is a neat catch, and moving voice IDs into a server allowlist closes a hole that most TTS demos leave open.
Two people is a good size. It's enough to catch one person's habit, and small enough that you don't end up averaging the pick into a voice nobody really likes. When the two of you split on a voice, how do you break the tie? Play it for someone outside the team, or go with whoever owns the script?
Usually whoever owns the script makes the call, since they know where it will be heard. If it's still close, we play both takes to someone outside the team with the names hidden.
Letting the script owner decide makes sense, since they know whether it will play on a phone speaker or in headphones. A blind listen from someone outside the team is a clean tie-break too, because they have no favourite yet. Thanks for answering all my questions on this one.
DemoVexa
Two numbers instead of one is the part most voice comparisons miss. A voice that starts fast but finishes slowly feels fine in a chat and drags in a long narration. I'd add two things to the review step. First, match loudness before comparing takes, because the louder of two takes almost always sounds better on a quick listen, whatever the voice. Second, hide the voice name while listening and only reveal it after you've rated the row. Otherwise it's easy to end up confirming the voice you already liked on the provider's demo page. The getAll check for exactly one voicePreset is a neat catch, and moving voice IDs into a server allowlist closes a hole that most TTS demos leave open.