Last month’s entry focused on music – how it’s developed and why. I mentioned how music is dynamic in nature, and how its versatility allows it to be used almost anywhere. This month, we’ll examine a different kind of audio asset which is almost like music’s alter-ego: voices.
Voices are tightly coupled with the game, because they’re literally based on the plot. Each line of dialogue corresponds with the voice of a character. In The Silver Lining, this will either be the Narrator, Graham, or some other character that speaks in a scene. The characters do not have to be human to have voices, they just need to speak English. Bird chirping, dog barking and other non-language sounds are considered SFX (i.e. sound effects) rather than voices, and will be discussed in a future entry.
While voices are basically the audio representation of words on a page, their design is much more elaborate. Technically, anyone able to read this sentence is capable of voicing a character. But voicing a character is more than just reading lines, it’s acting them. The expression, tone, accent, speed, clarity and intensity are all factors that come into play when choosing the voice for a character. This entry will describe how TSL’s voices were done.
Reference Material
Unlike other assets, most of the voices used in TSL were not based on how characters sounded in the previous games. While voices only existed in KQV and up, they varied so much in quality, that even to this day people comment on how annoying some voices were to listen to while playing. So, in order to choose the correct voice for a character, we had to look no further than the script. For example, take the following script excerpt from Chapter 1 of TSL with Graham & Hassan:
GRAHAM: Do you have any idea who that stranger was who cast that dark spell on both Alexander and Rosella?
HASSAN: Stranger? Sorry, friend, but the crowd was so riled up, I couldn’t see a thing. I frankly don’t want to get involved with something that nasty.
GRAHAM: Even if it might mean restoring the health of your king?
HASSAN: Now look, don’t get me wrong, but I don’t stick my head where it don’t belong. My way is the way of the sea and the tides. Mysterious strangers… I’m not going there. It would either take someone real brave or real stupid to go after a fella like that.
GRAHAM: Or, someone deeply disturbed by the status of the King and his sister, huh?
HASSAN: You do whatcha have to do. We all have our place in life. Like I said, mine is the sea.
Just from that small snippit, you can get a general idea of Hassan’s personality and casual way of speaking. Compare that with the dialogue of a completely different character, Lord Azure:
GRAHAM: Could I humbly ask a meeting with such honorable Oracle?
AZURE: Your request is declined. The disgrace over your family is not one of our concerns.
GRAHAM: It is of a great importance to all of our lands. The King himself is under a spell!
ARIEL: Azure, my love, let us not forget what the King did for our daughter.
AZURE: And he was rewarded.
ARIEL: I know. However, if King Alexander was great enough to be part of one of our sacred prophecies... as you said the prophecies foretold... perhaps we should not deny the possibility to aid him in his hour of need.
Azure thinks about it.
ARIEL:(CONT’D) There are prophecies yet to be deciphered. If the King fulfilled one, could he not fulfill another?
AZURE: I find that highly unlikely, but you have proved to be right on a few occasions. Human, you may see the Oracle this one time.
GRAHAM: Thank you, kind Lord.
AZURE: My guards shall escort you.
It is clear from the outset that Lord Azure has no real desire to help Graham, and has a slight tone of arrogance in the way he speaks. The formality of the words also indicates that the voice should be much more refined versus the ruggedness of Hassan's.
Auditions
While it’s great that the script can speak for itself figuratively, unfortunately it can’t speak for itself literally. The only way to find the correct voice for the part is to recruit people and hear how they sound reading the lines. But, with TSL’s script being over 2,000 pages long and our promise to not spoil the storyline for everyone, how could we possibly find the right voice for each character?
Through auditions, of course! Back in January 2004, we held in-person auditions in New York City, where dozens of talented people came to try-out. Most had the advantage of acting experience, and thus were able to naturally turn words on a page into natural speech. The way we did this was by having them read 1 of 6 character monologues (a.k.a. sides). Based on their tone, accent, annunciation and expression, many of the major character roles were cast.
A few months later, we held online auditions which were open to anyone in the world to try-out. They were given the exact same 6 sides to read – 3 male, 3 female. While about 30 people came to try out in-person, over 300 entries were sent in by people trying-out online. The same 6 audition sides were used in both auditions. The following is what the side looked like for “Male – Heroic”:
I had been King of Daventry for a few years, but I was lonely. My advisors would hold these festivals where princesses from everywhere would come to meet me, hoping I would be interested in one of them. I never was, however, even though they generally seemed very nice. And then one lonely night, the mirror showed me an image of this beautiful woman. What intrigued me most was what I saw within her eyes... I knew I had to search for her. And so I did. I traveled through the lands of Kolyma, and into this strange magical world beyond it, until I found the tower where my Valanice was imprisoned.
No prizes for guessing which character actually says these lines in-game. But, it is interesting to hear the actual audition for the role of King Graham. In the video below, you can hear Jason Victor's reading for yourself and also hear some lines for “Male – Evil” (a.k.a. Shadrack):
Recordings
Once the voice cast has been selected, the next step is to record all their lines. The live actors were all recorded in-house (literally!) by our Voice Over Director, Mike Fortunato. A microphone is hooked up to a computer and digitally records the audio to a voice editing program. First, several seconds of silence is recorded to measure the ambient static noise in the room (i.e. the whizzing of computer fans, noise from outside traffic, etc). By isolating the noise, it is easy to run a filter to remove it from the recording later on, producing a crystal clear recording of just the character voice.
Figure 1: Screenshot of a Voice Recording
Next, a few test recordings are done to ensure the volume is set correctly, based on the natural volume of the actor’s voice, and their proximity to the microphone. In other words, someone that’s naturally loud when they speak should be sitting slightly farther away and/or the microphone audio gain sensitivity should be reduced accordingly.
It’s rare for someone to read the lines perfectly in one take, so sometimes a few takes of each sentence is necessary. Mike will provide the actor with some direction, indicating if something should be emphasized differently, said slower, faster, etc. While it is tedious to say the same line over and over again, it is great to have different variations to compare and contrast for the next phase.
For online auditions, the process is similar but the recordings were first done by the person and then sent to Project Director, César Bittar, to provide feedback. The online roles were for minor game characters, and thus required less direction. Fewer lines and less direction means the recordings were able to be produced much quicker. The only disadvantage with online auditions is that the recording quality is not 100% controlled. Usually, by eliminating the static noise and applying other digital filters the voice quality remained consistent from person to person.
Editing
The main purpose of the editing phase is to correctly choose the best recordings to use, and to organize them correctly for in-game usage. The best take of each line is spliced together into one single file. This file will be clean of any other superfluous audio (i.e. reading mistakes, general chit-chat, Mike, etc.). Here's just one of the many outtakes that still makes me chuckle:
Graham recording (outtake):
Then, the recording is split up by scene. Some characters only speak in one scene, but other characters (like Graham and the Narrator) speak in many. Next, they’re broken down by conversation topic, and then by each line of dialogue within that topic. What you end up with are many audio files in the format of:
When we originally created this naming structure, we thought it would be sufficient for every line of dialogue; however it later turned out that it doesn’t quite accommodate for dialogues that are wordy. This occurs when text blocks are so large they fill the entire dialogue area on-screen and get truncated, because they overflow the visible region. This issue is more clearly shown in the screenshot below:
I’ll explain in a future journal entry how we managed to modify the game to accommodate this problem, but audio-wise it more or less meant that lengthy audio files had to be split up and separated into smaller files.
Knowing where and when to split dialogue is still somewhat subjective. Lines that are too short pass by too quickly to remember what has been said. Lines that are too long make the game too boring to sit through. The best length is usually a sentence or two per line of dialogue, depending on how quickly or slowly the text is read.
Sound Effects & Normalizing
The final stage in voice design is a combination of matching a voice to its environment, and making sure it's heard loud and clear. If Graham is in a large hall or cave, you would expect his voice to echo or reverberate accordingly. The way to adapt a voice to its surroundings is to use sound filters that mimic spatial acoustics, depending on the scene.
Normalizing volume makes the voice roughly the same volume level as any other voice without distorting it. It is essentially the same concept as mastering music, however not all sentences should be at the same volume and not all characters speak at the same volume level. So, just like adding filters, normalizing is a subjective process done on scene-by-scene, and sometimes line-by-line basis.
This stage is usually done post-production because it requires visuals, music/ambience, and basic scripting to be completed first. Most of the time only normalizing is required, but sound filters provide an extra level of realism that you don’t really notice until you hear it before and after.
To hear the impact that sound effects make, compare the following two sound clips. The first is a recording of the Oracle's voice in Chapter 1, while the second is the same with dialogue said with sound effects and filters applied to it.
Recording:
Recording with SFX:
If there's one thing you've learned from this entry, it's that voicing a character is much more than simply reading lines. It's almost like giving a character life, because the visuals can only convey so much. The voice is what allows the personality of a character to really shine through, when facial expressions or animations are lacking in dialogue. Sometimes facial expressions and animations are based on the voice itself, since a character's vocal tone determines when a character is happy/sad/angry/scared/etc.
The process of recording voices can be arduous. It takes time & patience, to say and listen to the same line over and over, in order to ensure the emotion from the script is properly conveyed. It also takes skill & timing, in order to know how something should be said while taking into consideration the character’s personality and motivation. While, technically, voices are as optional as music is to the game, they add a dimension of reality which improves the overall gaming experience. Next month, we’ll switch gears a bit and discuss the glue that keeps everything together. No, that’s not code for love; but in my opinion, it certainly does come close!
