First thing Thursday morning, Bungie Audio Director Marty O'Donnell and Audio Lead Jay Weinland held a presentation on their (newly-awarded) work for Halo 2. O'Donnell presaged the talk with a disclaimer. There are two sides to audio, he said: production and implementation. If you want to learn about production - composing, recording, directing actors - read a book. You can learn about that anywhere. In contrast, there aren't many resources on implementation - so that's what he was there to demonstrate.
"The main Bungie approach to games," O'Donnell said, "is this is entertainment. When someone sits down, you want to keep him entertained the whole time." This starts from the moment the console is powered up; over the corporate logos, a custom piece by Steve Vai leads into the game's opening theme. "The music at the beginning of the game," O'Donnell continued, "is the overture." It establishes a theme, to be used throughout the game. From the title screen, O'Donnell pressed "start;" as the game loaded, a motivated piece of music began to play against the Halo 2 logo.
O'Donnell explained he never wants to see the word "Loading": It's not entertaining. You always want the player to feel like something exciting is about to happen. "I never want an excuse for someone to get up and leave the game, if possible." The key to that is flow. O'Donnell prefers to think of audio as a cohesive whole; he would rather not have any one piece stand out.
O'Donnell and Weinland use a proprietary system named "Guerrilla" to control all audio data and parameters; Guerrilla's interface consists of an elaborate, branching tree of topics in the left frame, and variables in the larger window to the right. As O'Donnell rifled through options, he explained the game's dynamic mixing system; all sounds are arranged in classes, with each class bussed to an aux track. For instance, all weapons fire will be grouped together; this allows the designers to raise or lower any individual component on demand. To demonstrate, he loaded a debug level that he dubbed "Audio Playground." He described the level as a sort of skyscraper where every room is covered in a different material type. Debug information scrolled on the upper-left, showing what sound files were playing, when.
Jay Weinland and Marty O'Donnell
O'Donnell then explained the difference, in their system, between an "actor" and an "actee;" an actor is an object doing the hitting, while an actee is the object getting hit. Any collision results in a sound reaction from each agent. To demonstrate, O'Donnell steered Master Chief to a boulder and pushed against it, causing a grinding noise. He then boarded a tank and spent several minutes driving over other vehicles, to show how the sounds triggered. On deciding which noises stand out, O'Donnell said he wants some things to be "stars;" to be heroes in the game. "Real-sounding is not good. It's just real. Real is boring. We want sound to be visceral ."
He showed how he can map sounds to particle effects, like a shell casing hitting the stone floor. He can cap the effect, setting how many play at once before the effect gets "graduated": starts skipping cues, to prevent overloading all of the sound channels with nothing but shell casing noises.
Halo 2, O'Donnell says, narrates in a first-person tense. Although at times the camera pulls back to allow a third-person view, as when driving a vehicle, the idea is always to cheat a bit to make the player feel more involved. Usually when an object is close, it is in stereo; when it is at a distance, it goes to mono. When driving in the third person - they figured whenever you get into a car, you want to see what you're driving; it's neater that way - the sound of the vehicle remains in stereo, somewhere between how it would sound when driving and what you might hear from the outside. The mono channel gets pegged to the left and right sides of the screen, and a routine is passed on it to make sure it always stays in-phase.
Voicing the Characters
Next, O'Donnell showed the sound banks for all of the characters in the game. Halo 2 has three-hundred eighty-four AI states, with an average of one-hundred forty for each character. Each state has somewhere between two and ten-plus voice samples, depending on the actors' improvisations, meaning each character has an enormous amount of potential cues. O'Donnell scanned through the cues - several hundred - for the "Sgt. Gruff" character, until he reached the curses. Each had a frequency assigned to it: " You split-chinned bastard! " might be tagged "somewhat," while " Get back in the bowl, puke! " might be "rarely" and " You giant no-good pile of hoo-hah! " "almost_never." O'Donnell noted that although Orlando Jones, who voiced the "Sgt. Cautious" character, was great with improvisation, they had to cut half of his material because he managed to work the word "fuck" into every other take.
Once the cues are imported, they are assigned to "skip fractions," that determine how often the sound is skipped, rather than played, when the engine tries to trigger it. At times, the rarest sound clips break the fourth wall; as when, after the player has abandoned the game for a long period, perhaps to go to the kitchen, then returned, Sgt. Cautious might ask " So, did you get me anything? " The idea is, maybe one player in a few thousand will happen be in exactly the right place, at the right time, for that one bizarre clip to play - so when he tries to tell people on the Internet about it, "there's no one going to believe they heard it."
Halo 2 uses what O'Donnell calls "Quantum Music;" he uses this term instead of "dynamic music," as does not feel that the player should understand he's in control; then it becomes a music game. Events trigger alt loops; each piece has a natural end, while the middle can last however long. If a piece plays for too long on one loop without the player doing anything to advance the music, it starts to fade out, to avoid getting annoying. O'Donnell loaded an example piece in a wave editor, to show how the piece was chopped into loops, with each loop a few measures long. "This is a good end," he pointed out, visually. "This is a good transition." The idea is that they have control over beginnings, middles, and ends of pieces; the middles are variable.
O'Donnell played a "glue" piece that consisted mostly of a C-pad loop, and was used as a transition between any two other tunes. He had a palette of elements to lay over the top of the pads, like a drum loop or "hits," cued to dramatic actions in the game. He showed the dramatic effect it has when he fades out the pads, then keeps the drums going for a moment. This brought him to his next point: "I love to have music that doesn't stop just because a cinematic starts." He showed a clip of the game, as the player triggers the same cutscene at different times in the loop. In the first case, the background music ends and transitions to score about five seconds into the cutscene, blending the atmosphere well; in the second, the original piece continued to play through almost the entire cinematic, ending maybe a half a second before the scene did. This worked just as well as the earlier example.
"I want people to have a great experience however they play the game," O'Donnell concluded. "My philosophy is... the ear does not blink." Any interruption, any abrupt change, any conscious notion that the player is in control of the music, will bring the player out of the experience.