Skip to content
MageCats

Dubbing notebook

How to pick one voice out of hundreds

Character sheets, auditions with the character's own lines and stress tests: how we cast a dub with hundreds of roles.

Published · 6 min read

The San Andreas dub has almost 500 voices. The Dota 2 one, about 130. Behind each of them is a casting decision, and most of those decisions go unseen: nobody notices pedestrian number two hundred until he sounds exactly like pedestrian number one hundred.

This article explains how we choose voices when the roles run into the hundreds, and which mistakes taught us to do it better. A note on the craft, in the interest of honesty: a good part of our cast are directed synthetic voices, and the casting works just as it would with actors: you write a part, you hold auditions and you listen.

First, the part

Before looking for a voice, you need to know what you are looking for. For every character we write a short sheet with four items:

  • A specific apparent age. Not “adult” but “about 32”.
  • A specific accent. Not “Latino” but Caribbean, northern Mexican ranchero, Chicano, Rioplatense, Paisa, Chilean…
  • A vocal trait. Raspy, nasal, high-pitched, smoker’s, drunk, nerdy, hot-headed.
  • Temperament. How the character reacts: curt, preachy, paranoid, warm.

The age that matters is the apparent one, not the one the game’s backstory gives. In San Andreas, the protagonist is twenty-four according to his biography, but the way he talks and everything he has lived through call for a voice in its thirties. Casting follows what the player perceives.

What the protagonist taught us

CJ was the hardest role to cast, and the one that taught us the most.

The first three rounds failed for a simple reason: we were looking for a “young” voice. The young voices sounded like students or like the announcer in a TV commercial, and CJ is neither. The right sheet turned out to be an adult Afro-Caribbean man, about thirty-two, laid-back and a little raspy. With that sheet, the voice turned up almost at once.

The second lesson came after hearing him in the game. One of the chosen voices sounded “passive, flat”. The problem wasn’t the voice: it was the direction. The acting notes said «cansado» (“tired”) and «resignado» (“resigned”), and the voice did as it was told. CJ is a dry protagonist, but he is not a man without energy. We changed the notes and the character came back to life.

Since then, we check the direction before we blame the voice.

Auditions with the character’s own lines

Every audition uses the character’s own lines, taken from his already adapted script. Never a generic test text.

There is a practical reason for this. With a generic text, like «Hola, ¿cómo estás? Hoy es un buen día» (“Hi, how are you? It’s a nice day today”), nearly every voice sounds fine. With a real line from the character, say a whispered threat or a shouted insult, it becomes obvious right away who can carry the part and who can’t.

We also learned that the sheets can’t come out of a template. In one round of voices for the radio, we wrote the sheets for 42 call-in listeners with almost the same words and the same test text. The result was a parade of near-identical voices: the same man and the same woman repeated across the whole group. Now we spread out accents, ages and traits so that no two listeners share the same combination.

The stress test

Some characters ask for more than a good voice: they ask for stamina. Catalina, in San Andreas, goes from tenderness to fury within a single line, laughs like a madwoman and threatens to kill you in the next. A voice that sounds good on a calm line can fall apart in those shifts.

For those roles we run a stress test: the same voice reads five very different lines, of fury, tenderness, crying, unhinged laughter and menace. Then we listen, and we also measure: how much the volume rises from tenderness to fury, and how much the pitch moves. A voice that says everything the same way doesn’t pass.

Measuring doesn’t replace the ear, but it helps it. Other things, like how dramatic or how effeminate a voice sounds, no number can capture: there, only listening counts.

Accents, from native speakers

A strong foreign accent comes out better from a native voice than from an acting note. We have confirmed this several times.

When a character speaks Spanish with a German, Russian or Chinese accent, we look for voices of native speakers of those languages who speak Spanish. In Dota 2, for a heroine with a German accent, we compared a voice designed with the note “German accent” against a native German voice. The native one won hands down: the note produces an imitation accent; the native speaker produces the real thing.

The same goes for Latin American accents. The voices that do Caribbean best are Caribbean.

What we never do

There are two casting rules that are not up for negotiation:

  1. We never clone or imitate the original actor’s voice. Every character gets a new voice. The dub is a performance of its own, not a copy.
  2. We never rule out a voice because of what it “seems” to be. A name or a title doesn’t tell you where a voice came from. A voice is only ruled out on verifiable evidence, such as a takedown or a complaint. The same applies the other way round: a voice doesn’t get in just for having a good name in the catalog.

That second rule also protects the casting from itself. The best voices for specific characters tend to be little known or to come from another language. A filter that is too narrow hides them. That’s why we run broad searches, in several languages and with several sort orders, before proposing candidates.

How the decision is made

The candidates go onto a listening sheet: a page with each role’s auditions side by side, which can be played from a phone. In comparison tests, option A is always the voice already in the dub, so the comparison is fair: the new voice has to beat the current one, not nothing.

The decision is always made by ear. There have been cases where the measurements pointed to one option and listening chose another. Listening wins: measurements guide, the ear decides.

After the casting

Choosing the voice isn’t the end. A voice that sounds good in an audition can get lost inside the game, with music, engines and gunfire all around. When a character “can’t be heard” even though he is as loud as the original, he is almost always missing brightness: energy in the high frequencies. We measure it against the original voice and correct it with EQ before matching the volume.

And if, once in the game, a character doesn’t convince us, we recast. In San Andreas we have changed the voice of several main characters after hearing them in play. It’s more work, but a character with the wrong voice is heard for forty hours.

← Back to Dubbing notebook

Follow us on

Dubbed clips, live streams and a heads-up on every new release.