[Feedback] GLaDOS voice

[espen180]

It will forever be a mystery.

Anyhoo, I DID "aqcuire" this program, but there was a fault with the "Aperture Science Pirated Software Registry Ownership Licence Entry Application", so I couldn't get it working.

[nickworks]

The problem with text-to-speech is that it lacks the necessary pitch to sound realistic.

When I listen to text-to-speech, I have a much harder time listening to what is being said. We can listen more passively if the speech has tone, pitch, and inflection.

Anyways, I seem to remember the commentary talking about mixing two samples: the actress's and some computer generated audio.

I'm guessing that they are describing what's called a Vocoder. If you listen to music and the singer's voice sounds way too smooth and digital (I think Cher has a song like this that was pretty popular... 5 years ago?), that's thanks to a Vocoder.

From what I understand, it uses one sample to control the pitch of another sample (and also mix the two signals, if you'd like). The way I would do it is have an actor deliver their lines monotone. Then create a signal that you control the pitch/note of. Use the signal to assign appropriate pitch to the dialog. The vocoder settings should let you decide the quality, wet/dry mix, and more.

All of this is easy to do in Fruity Loops.

Cheers,
Nick

[segfaultii]

nickworks wrote:

All of this is easy to do in Fruity Loops.

Well, post an example! I'm eager to hear the fruity loops edited voice.

[Player1]

That Melodyne tutorial posted above was really good. Here's an example of what I achieved using it: turret rant.

I used the demo version and simply used fraps to record the sound directly out of Melodyne as it played it back.

This took all of 5 minutes or so. I'm sure it can get much better if you massage the input to the voice synth and do some more pitch editing in Melodyne.

EDIT: Changed to MP3 instead of WAV to save the rainforest!

[segfaultii]

Nice, thats a good example.
To enhance the likeness to GLaDOS, slow the track down a bit and try some more variation in the pitch.

I found another TTL service that has a somewhat deeper female voice: http://cepstral.com/demos/ (use the callie voice)
Might be worth a try.

[Player1]

Wow this is fun. I've played around with it some more. There's no doubt that the AT&T reseach thing is the best one, and the most GlaDOS sounding one as well. Unfortunately their legal section explicitly lists game levels and non-profit use as no-go, so we probably can't use this for real :/

I did some more examples and experimented a bit with speed and manually editing pitch of selected words. I also added some reverb to simulate what the HL2 engine does to the samples in-game (although the specific effects used are very "signature" so the effect isn't quite the same:

glados1.mp3
glados2.mp3
glados3.mp3

[nickworks]

segfaultii wrote:

Well, post an example! I'm eager to hear the fruity loops edited voice.

Well there are dozens of ways to do this, depending on your goals. For instance, your AI character might have a completely different personality. I played around with my FX processor on my mixer and came up with some interesting results.

But about the Fruity Loops vocoder. I decided to try it out. Now, I'm not very good at it, so it doesn't sound perfect, but if you listen you can hear where I used a second signal to control the pitch of my voice. Obviously there's a lot of refining that could be done with the acting, the audio cleanup, and the actual settings of the vocoder (wet/dry mix, frequency, # of bands, the modulating signal) and the other filters I ran it through.

You can really clean up a sound of an audio file by passing it through a multiband compressor, or (depending on the game-play environment) a delay. I wouldn't use a phaser or flanger as I saw some people trying. Those do something completely different.

In the file: Unedited voice first, then the vocoded.

http://myaperturelabs.com/vocoded_npattison.mp3

It's okay, but a little too distorted.

[espen180]

Yeah, I've tried every single program I could find, and it won't work because of three things:

1) The keygen doesn't work.

2) The demo doesn't allow saving.

3) The program costs $999.99

I just found a freeware program called Sound Indepth 1.1. I'm gonna try that.

[Rivid31]

I have "aquired" Melodyne as well, works fine for me. It seems a little buggy of a program though, maybe it's just my computer, but when you play any sound in it, it cuts in and out every now and then... :shrug:

But, I can at least save my wavs, and they sound normal when played in another program.

[espen180]

Dude, what keygen did you use? The one I downloaded from demonoid couldn't create the license number.

[hamsteralliance]

I think I have pretty good GLaDOS voice down. I'm not using text to speech but actually recording the dialog myself and processing it with Melodyne and Soundforge.

UPDATE 1: Here's a newer example - I think it sounds better: 02_useless_button.mp3 ( I'm certain that button doesn't do anything, it's just there for decoration.)

-----------------------------------------------------------------

Here's an example:
01_best.mp3 (Whoever said the best things in life are free, never had an Aperture Science Hand-Held Portal Device.)

There's a bit of plosive noise from that recording as I didn't have the pop-filter placed correctly but that has been corrected since.

I'll be making more tests tonight - so if you have any feedback for me on how to improve my GLaDOS-style, let me know.

My current questions:
1. Am I headed in the right direction?
2. Does it sound better than the text-to-speech attempts?
3. Should I talk in a higher pitch/inflection?

[Hober]

Sounds good, but the inflection isn't quite right. I'm not sure exactly how.

But the sound itself is really close. Maybe a tad to "tinny"

[hamsteralliance]

Thanks for the feedback, Hober.

Another try, trying to add more roundness (femininity/cuteness?) and emotion to my voice. Kind of sounds like a turret.

04_look_over_there.mp3 (Hey, look over there! It's your best friend the companion cube. Oh, nevermind, it was just a picture of a companion cube. Not your companion cube...You incinerated that one. )

[Player1]

It's fairly good, although a bit too tinny and a bit like coming through a telephone.

To me it sounds like your formants have been raised too high. It sounds more like the turrets than GLaDOS.

[NocturnalGhost]

I think your first one sounds closest. Your technique seems very good, but you need to work on your pronunciation.

Listen, for example, to how you say the word "Portal" compared to how GLaDOS says it. You need to sound more British. Pronounce your T's more, etc.

This is one of the closest i've heard so far though. Nice work.