Reply To: Mandelbrot Set: Laura’s Revenge
Home › Forums › JoCo Projects › Mandelbrot Set: Laura’s Revenge › Reply To: Mandelbrot Set: Laura’s Revenge
I was testing out the UltraStar-reading code I just added to my robot choir program, and I made these recordings, which might be useful, but are by no means obligatory. It’s The Future Soon sung by five Macintalk voices — three which are supposed to sound like human females but of course still sound computer generated, and two which deliberately sound like robots:
Humanoid females:
Robotoid Robots:
Of course a human voice would be better for most of the song, but I was thinking it could fade into the humanoid voice during ‘As I piece by piece replace myself/And the steel and circuits will make me whole’ to show that the narrator is becoming more of a cyborg, and then maybe go abruptly go to the full robot voice at ‘Hearing the whir of the servos inside’. Or whatever the mixer thinks is best. I like how the word ‘whir’ sounds quite a bit like whirring servos when the robots sing it.
I can easily change the lyrics to gender-bent ones once they’ve been decided upon, or change the octave, or key (if I understand how keys work), or voice; making the recording is just a couple of clicks.
I used the Future Soon file from Spektugalo’s UltraStar countdown and transformed it into speech commands. There are a few weak and strange things I still need to engineer away. I think the lengths of pauses between lines are not quite right; the UltraStar file is kind of ambiguous at newlines (there a – followed by what seems to be two note positions… I just made it ignore the time between those two positions because otherwise the gaps were way too long.) I can’t run UltraStar on my Mac (it’s a G4… supposedly UltraStar Deluxe and NG can be compiled for G4 but I never got it to work) so I can’t experiment with it to figure out the format until I get my shiny new Intel-based MacBook Pro some time this week. I also need to improve the way my program decides how much time to give to different phonemes… sometimes it lllingers too long on a consonant.
I noticed that Vicki and Victoria’s recordings run for around 4 minutes, close to the length of the original, while the other voices go for 4:29. This is not too surprising, since Vicki and Victoria are newer voices in a newer format (I think Victoria came out in Mac OS 8, and Vicki in Mac OS 9, while the others have been around until at least System 7) and so are more likely to be able to do exactly what I tell them. The TUNE format I’m using to tell them the tune to sing only came out in Mac OS X 10.2. [ETA: Depite the name, it’s not really intended to make the computer sing; it’s supposed to be for people who want to fine-tune the intonation of normal speech (if you’re a curious geek, check out the Repeat After Me program in the developer tools. You type in a phrase, record yourself saying it, and it spits out TUNE commands to make the speech synthesiser intone the phrase exactly the way you did. This, combined with my lack of singing ability, is basically what gave me the idea of using the TUNE format to sing. I met the man who wrote Repeat After Me in 2004; in his spare time he is a magician, and he looks like Gandalf.)]
Anyone with a Mac with Mac OS X 10.2 or later can experiment with other voices by opening this file in any text editor and choosing ‘Start Speaking’ in the ‘Speech’ submenu of the Edit menu. You can change the voice in System Preferences, or use some program which gives you a menu of voices directly if you want to more easily experiment. I just checked they sounded okay in the default voice and then experimented with different voices using the ‘convert text to sound file’ action in Automator, since I needed to make sound files of them anyway.