When I was asked to develop the artificial intelligence (A.I.) for the squad for Star Wars Republic Commando™, one of my earliest requests of the design team was for a large suite of dialogue lines to be written to help the characters feel alive, intelligent, and entertaining. I knew that no matter how “intelligent” I could make the A.I., my efforts would be in vain if the A.I.-controlled characters couldn’t express what they were thinking to the player through voice and animation.
I started working on voice design early by generating a list of some of the situations I wanted the A.I. to be able to recognize and speak about. Each situation was given a unique cue name or I.D., such as “KilledDroid” or “AssistingLeader,” that could be used by both the programmers and sound designers to identify the situation and group the appropriate sounds.
My initial list contained about 50 cues for situations that would provoke the squad to speak. There were cues for being hit or killed by different weapons, for detecting different enemy types, for responding to the player’s commands, etc. Each character type had its own set of sounds indexed by cues.
To avoid the problem many games have of A.I. voice being repetitive and to make sure that the player wouldn’t tire of a small number of A.I. voice lines, we generally wrote 10 variations for each cue. With three non-player characters in the squad, 50 cues, and 10 variations per cue, we had about 1500 lines, and we were just getting started!
Even with all that dialogue, it seemed every week we found a new situation where we wanted a new voice cue. We paid special attention to having the squad recognize and respond to strange player behavior, such as the player countermanding their own orders (cue name: PlayerSchizo), firing their weapon at nothing or at their teammates (cue name: PlayerShootingGhosts), killing themselves (PlayerKilledSelf), getting in the squad’s way (PlayerOutOfMyWay), etc.
Several months later, we had more than doubled the original number of cues and lines, and though the squad was sounding intelligent, something just wasn’t right. Each of the squad members sounded too similar, and focus testers could not tell the difference between them. We took the difficult but necessary step to recast the squad and re-record all of their dialogue in the game. We also refocused our writing efforts to better tailor the lines to express the personalities of the individual squad members. For example, Fixer, our player’s right-hand man, often suggests tactics or helpful advice with his cues. Scorch, the demolition expert and wiseacre, jokes around. Sev, the bloodthirsty one, is dead serious.
Our sound and voice designers made a Herculean effort to record, edit, and wire all the new dialogue in a short amount of time, and the payoff was fantastic. The new character voices and lines helped immensely, and I was almost satisfied … but not quite. Although the new dialogue was entertaining and added greatly to the illusion of squad intelligence, I wanted a group personality evident as well as the individual personalities. So, one of my recent additions was a squad banter system that allowed us to replace a voice cue on a specific level with a mini-conversation composed of lines from multiple squad members.
Finally, at the end of the project, we had thousands of lines written for more than 180 different cues. Despite the hurdles along the way, I’m very pleased with the results of the team’s efforts in dialogue. Even though we ought to be sick of the dialogue by now, since we hear it every day, the squad is intelligent, alive, and entertaining because of it, and I can’t wait to see how people respond to our efforts.
