What a voice casting poll taught me

The biggest thing my voice casting test taught me is that a fair comparison makes the decision clearer but not automatic. For Duma Meow, I had every one of the four cats say the same line, "Ham chơi!", in every one of the four candidate voices, in a balanced order, and I still haven't made the final pick.

Here's how the test worked and what the process showed, so you can run something similar for your own characters.

The setup

Duma Meow is a Vietnamese cartoon comedy about four cats: Maelis (deadpan), Corin (flustered), Seraph (wildcard) and Luna (always prepared). The voices are generated in Vietnamese with ElevenLabs v3.

Each cat has a provisional voice:

• Maelis: Lily
• Corin: Liam
• Seraph: Jessica
• Luna: Laura

"Provisional" is the key word. Those pairings were a starting point, not a final decision, so I wanted a way to test whether a different voice might fit a cat better.

Why one line?

Using the same line for every take removes one big variable. If each cat said a different line, I'd be judging the line as much as the voice. With one short line, the only things changing are the voice and the character on screen.

Ham chơi (say "hahm choy") roughly means being too into playing or fun, neglecting what you should be doing. It's short, expressive and easy to compare. You can hear it in the ham chơi voice test Short and read more in ham chơi meaning.

Why a Latin square?

Four cats times four voices gives 16 combinations, so the test has 16 turns. The order of those turns matters more than you might think. If one voice always comes first, it might seem fresher, or it might set the bar that the others are compared to.

A Latin square is a simple grid arrangement that balances this out. Each voice appears once for each cat, and the order is arranged so no voice is always first or last. It doesn't remove bias entirely, but it stops the order from quietly deciding the result.

What the process taught me

These are lessons about the method, and they'd apply to any casting test:

• Hearing every combination changes your assumptions. Once you've heard a voice on a different character, the original pairing stops feeling inevitable.
• The character on screen affects the voice. The same take can feel different depending on whose face it's attached to.
• One line isn't the whole story. A short exclamation tests energy and tone, but not how a voice handles a longer or quieter line.
• Fairness slows you down. A balanced test gives you more to think about, not a single obvious winner.
• It's okay to leave it open. A provisional cast lets production continue while the decision matures.

Polling yourself vs polling others

A test like this can be judged by one person, which is what a casting director does, or it can be shared with viewers to get reactions. Either way, the structure is what makes the answers comparable. If you ask an audience, show them every combination in a balanced order, and ask a simple question, like which voice fits that cat best.

How to run your own casting test

• Pick one short, expressive line that your characters would plausibly say.
• List every character-voice combination.
• Balance the order with a Latin square or a simple rotation.
• Judge with the visuals, since voice and character work together.
• Write down quick notes after each turn, before you hear the next.
• Sleep on it before deciding.

Where things stand

As of now, the provisional voices are still in place and the final pick is still open. When it's made, the test will have done its job: giving every option a fair hearing. For more on the voice process, see how I cast voices for four cartoon cats and making a Vietnamese cartoon with AI voices.

If you're casting your own characters, record one short line in every voice for every character before you commit; the comparison will tell you more than any single audition.