Day 8 saw the arrival of two colleagues and one hefty tarantula.
Despite having to negotiate the Atlanta airport with speed, they made their connection to Santiago, where they obtained their Mountain Names: Iden and Jackke.
Nobody captured the exact moment they alighted from their servicio especial van, and they’re extremely asleep right now, so you’ll have to trust us: they’re here.
Today we crammed in quite a lot of science and engineering before shutting the instrument down for the night before its Friday morning trip up to the telescope platform. We also wrap it up to prevent heating and intrusion by dust.
Anyway, all that means: tomorrow we’re doing the thing! This means we wake up for breakfast, make approximately 100 trips back and forth between the summit and the almost-summit where the cleanroom is, plug a lot of things into other things, and then do astronomy until it’s time for breakfast again.
As the only graduate student present who has been on a MagAO Classic run, it fell to me to implement the speech-synthesis ops concept that Jared has long coveted for MagAO-X. (Only so many people can crowd around MegaDesk, but everyone in the control room can hear “loop is closed.”)
Speech synthesis has changed a lot since Jared was a grad student, and professional-grade deep-learning models are surprisingly good these days. Of course, the budget for this was $0.00, so I used OpenTTS. OpenTTS lets you sample a wide variety of synthesis packages with a common API, and I had the container running on AOC (that’s our Adaptive optics Operator Computer) in a few minutes.
The next challenge was to hold auditions. For my auditions I imagined some things MagAO-X might want to say, and gave them to the various models: “Target changed. High-order loop closed. Data saving started. Loop is open. Smoke ’em if you got ’em.”
The previous-generation models were there for comparison, and sounded like absolute butt:
On the plus side, espeak was one of the only included engines with a non-US/non-UK English voice:
Island vibes aside, the synthesis was just not that good. I kept looking. This MaryTTS voice seemed all right:
Just one small problem: Maggieo-X is clearly not male. This voice clearly is. That simply won’t do. How about this from Larynx?
Not bad. But are we sure Maggieo speaks English?
I also ran the Dutch voices past Sebastiaan Haffert, local Dutchman:
Verdict?
“They’re all… very Flemish.”
“Is that bad?”
“No, they’re just very… soft.”
It’s true. The characteristic throaty “g” of the true Dutchman was let down by these synthetic voices. Just click the speaker icon on here to have Google’s model read the same to you. See?
Anyway, auditions continue, but I have a pretty good idea of who will be narrating our ops tomorrow night. Until then,
Song of the Day
Okay, maybe when I posted Bring Me To Life and then Jared had Korn it started a bit of a nu metal trend, but in the end it doesn’t really matter.
I don’t play many video games, mainly because I am usually in the business of telling computers what to do, and therefore really resent them telling me what to do. One recent exception was Sayonara Wild Hearts, which is incredible. More of an album-length interactive music video than a “game,” I guess, but anything that includes:
- firing heart-shaped projectiles
- from your motorcycle
- at a three-headed robot dog
- as a metaphor for overcoming heartbreak
- (narrated by Queen Latifah)
has to be good. And, of course, the music complements the v i b e s perfectly. (And is strongly inspired by CHVRCHES, no doubt.)
For some reason, nobody’s taken the unadulterated audio and overlaid it on gameplay for a ready-made music video, but if you can ignore the dings and bleeps this should give you some of the flavor.
Bonus Tarantula
Content warning: incredible arachnid. Click to tarantulate.
This little guy took up residence on Avalon’s door this evening, requiring XKID xpert Noah Swimmer to remove him.
