How To Create Glados Voice

GLaDOS voice generator

This tool lets you type in some text, and the server will generate a GLaDOS-like reading of the text.
Who/what is GLaDOS? The main antagonist in Portal, a video game by Valve. See the Portal Wiki article on GLaDOS for more information.

Why should I enter my email?

Because of the inefficient process used to generate voice lines (see below), it takes about ~10 minutes to generate a single line, and many people may use this generator at the same time. Thus, when you ask the server to generate a voice file, it adds your request to the end of the request queue. Your message will be synthesized when it moves up to the first position in the queue, which can take a while. Entering your email is a way for the server to notify you about the completion of the process.
If you do not enter any email, you will not be notified and will have to check the list of processed messages regularly.

Statistics

Messages processed: 122,498.
Messages in queue: 125,172.
Latest processed messages:
- "
  Neurotoxin would be very useful right now..
  " (Download)
- "
  You are so smart that snails should be worried of you outsmarting them.
  " (Download)
- "
  The quantum tunnel will open in three... two... one...
  " (Download)
- "
  hello, come to get the cake
  " (Download)
- "
  Why is this happening?
  " (Download)
- "
  Please stand clear of the doors or else, Goodbye.
  " (Download)
- "
  Beunos Dias, Mandy
  " (Download)
- "
  why do i exist
  " (Download)
- "
  Glados is you , not me
  " (Download)
- "
  Intrusion detected! Silent alarm has been activated. Neurotoxin will be dispatched shortly.
  " (Download)
Messages added to the queue recently:
- "
  Добрый день зрители. Как вы знаете много игр было сделано Valve corp. Поэтому мы вам расскажем когда была сделана та или иная ишра
  "
- "
  By the way, should I use glados text to speech from now on?
  "
- "
  weed actually makes you smarter. the brain cells cancel out
  "
- "
  You know your *high* when you exhibit the following characteristics: 1. you laugh uncontrollably because you realize how high you are and how high everyone else is and how everything else you wouldn't give a second thought about when your sober is now co
  "
- "
  The state I am in now. For the 5 people who have never gotten high, I will try to explain it. at first, before you are really high, you become really giddy and will laugh at the slightest thing. you also may remember random things and pointlessly share th
  "
- "
  siemka jestem glados lubiem fortnajt
  "
- "
  hieeee
  "
- "
  No. You're shitting me. You're telling me a ginger bred this man?
  "
- "
  Hail pasta
  "
- "
  essential flow
  "
The full list of processed messages (with downloads) is available here.
Web server is up, database server is up, synthesis server is down.

About GLaDOS's voice

GLaDOS was voiced by Ellen McLain, a professional voice actress. This tool is automated, thus cannot rely on professional voice acting; instead, it uses text-to-speech synthesis to generate a female voice, then distorts it to make it more GLaDOS-like. This process is quite long and obviously produces results of poorer quality than the voice you hear in Portal; do not expect Portal perfection.

How is the voice generated?

There are various tutorials around the web on how to do this. Here is such a tutorial (as a YouTube video):

The problem is that this method makes use of GUI (graphical user interface) software, which cannot be easily automated. What this tool does is, indeed, to automate them anyway, despite the shortcomings of such a method. Here is what happens when you enter a message in the form above:

The message gets queued in a database.
The queue is processed at regular interval through a script ran with a cron job.
Once the message arrives in front of the queue, the server generates the initial voice sound file using Festival with the nitech_us_slt_arctic_hts voice.
The server is running Linux, so it has to run a Windows virtual machine, and launch Melodyne inside the VM.
A Python script takes care of opening the file in Melodyne and to apply the method described in the tutorial, using shortcut keys when possible, or by taking screenshots to determine where to click on screen when there is no shortcut key to perform a certain action.
The Python script then exports the file as a 44,100 kHz 16-bit linear PCM .wav file
The file is uploaded back to the web server, which stores it and sends a notification email if an email address was provided..
The server moves on to the next message, if any.