Creating a Portal AI Voice
Professional singer and voice actress Ellen McLain provided the voice of the AI GLaDOS throughout Portal. Several steps were undertaken to achieve the specific sound heard in the game. According to Director's Commentary in the game, McLain listened to synthesized text-to-speech voice samples to give her the general idea of how to deliver her lines. McLain then performed her lines in a monotone, robot-like manner. The lines were digitally edited, with the "pitch constrained, pitch modulation suppressed, and the formant moved up". This process is achieved through pitch correction, commonly referred to as "autotune" (though Auto-Tune is a specific brand of pitch correction software). While pitch correction is widely used to create a distinctive musical sound and correct off-pitch singing, the process can be used simply to make a voice sound more robotic and synthesized without necessarily sounding musical.
Several commercially-available sound and music editing packages include pitch correction features, which can be used to achieve this sound.
To create a good-sounding AI voice, you will need a recording microphone and appropriate audio editing software. It is strongly recommended to use a high-quality microphone designed for recording voice or music - you will not get suitable audio quality out of a gaming headset's built-in microphone. Your audio editor will need a fully-featured pitch correction feature, with the ability to manually adjust the pitch of individual syllables.
Some software with pitch correction abilities:
- Antares Auto-Tune Evo (demo available)
- Melodyne Essential (formerly Melodyne Uno. Demo available)
- MAGIX Music Maker series (pitch correction is referred to as "Elastic Audio" in this software. Elastic audio is not featured in the demo)
- Image-Line FL Studio series (pitch correction is available through a plugin called "NewTone". It comes pre-installed with the "Signature" edition of FL Studio, or can be bought later)
You will want to record and save your lines as separate files as they will be heard in the game. So if a trigger activates a three-sentence line of dialogue, record those three sentences in one take and save them as their own file.
The first step is applying a basic pass of autotune. This will generally shift each syllable to the nearest semitone (preferably in the chromatic scale in this situation), and may reduce vocal vibrato (constant subtle fluctuation in pitch). From here, you may want to manually shift individual syllables or words up or down in pitch; this will create a more artificial sound and can also be used to emphasize certain parts of the line. It helps to listen through the original GLaDOS dialogue repeatedly to get a feel for when to change the pitch.
Finally, by increasing the Formant slightly, you can give the AI voice a slightly more "squeaky" sound which will enhance the artificial feel.
To do: Add screenshots
Load your audio file into melodyne by going to File > Import audio. Clean up the audio if you like.
Select all with CTRL + A (CMD + A on a mac), then select the pitch tool. Double click on the audio (it should all have a red highlight) to send it to the nearest note.
Select the Pitch Modulation tool from the pitch tool. Double click again on the audio, making sure everything is still selected. This will flatten the pitch and make the voice sound like GLaDOS.
Step 4 (optional)
Use the Formant tool and raise the bars up (or down) one octave.
The voice is now finished and ready to implement into a map with a
logic_choreographed_scene or other application.