Hacker News Reader

Comment by marcan_42 | Hacker News Reader

It was never supposed to be nonsense. To anyone playing in Japanese or presumably other phonetically consistent languages, it's obviously a sped up, slurred/somewhat mangled version of what the text is saying. I guarantee not a single Japanese player didn't notice this, or otherwise thinks it's pure gibberish. Japanese is not my native language and I noticed immediately; any native would have too.

The problem is that English is a phonetically inconsistent language, with a massive number of rules required to even begin to approximate the mapping from text to phonemes (and zillions of exceptions). So this kind of really dumb TTS not intended to be actually intelligible doesn't work at all in English. And so it sounds like actual nonsense.

View on HN

It did work in English though. Maybe not as well, but I do remember noticing in the original Animal Crossing that there was some correlation between the sounds and the text. Even if it was just the timing and intonation, it could definitely be understood that the lines were actually being spoken in some form. I don't think it was TTS because it was too accurate in a really subtle way.

Or maybe it was and they tweaked it well enough to work. It's been a while, my memory could be off. I don't know if it would have been prohibitive on the GameCube to have audio of every line (there were a lot), but I wouldn't put it past them to have done so.

It's TTS. They absolutely don't go dubbing around every line for this, that'd be insane. That's the whole point of this system, to provide some fun sounding audio for the lines without having to actually dub them. You can tell because they speak your island name and your own name exactly the same way as the rest of the text, accurately. And because it's pretty monotone and consistent.

Besides, can you imagine voice actors dubbing this stuff in this kind of voice line by line? They'd go insane.

Yeah that makes sense. I guess I figured they could have processed the spoken audio, not that the actors would have actually talked like that. I was just fooled by the fact that their TTS sounds more passably human than any other I've heard (in an abstract way).

This technique predated Animal Crossing, though.

Banjo-Kazooie [1] was the first game on the N64 to use what Animal Crossing terms "Animalese/Bebebese" [2], and their intention was never to build a TTS engine. The first Banjo was released in 1998, while Dobutsu no Mori (Animal Crossing) for N64 didn't come out until 2001.

Nintendo was definitely talking with Rareware at the time and they exchanged ideas and techniques on game engine design, platformer mechanics, etc. Interviews from the Rare side admitted this (I'll need to dig up some sources to include here).

I'm curious if Nintendo picked up the Animal Crossing Bebebese voices directly from Banjo-Kazooie.

[1] https://www.youtube.com/watch?v=9ZE5A3DbHDk (Actually a video of the year 2000 sequel, Banjo-Tooie, but this is a better example of the same voice engine using various voices.)

[2] https://animalcrossing.fandom.com/wiki/Language#Bebebese

I've played it in both Japanese and in English, and while Japanese is more phonetically consistent I don't understand how people could miss it in English either. You can tell right at the beginning when Rover says your name back to you. Even if it's usually difficult to make out, occasionally you should notice it's not totally random, especially when you start playing around with giving the villagers catchphrases.

This reminds me of the game killer7, where some NPCs would speak in a partially distorted voice. However in that game it always quite interesting listening to the TTS because the script an NPC is reading seems to be a paraphrased from the subtitle (and maybe partially complete gibberish?) and it feels like you’re fading in and out of understanding the language as they speak [1], which is in fitting with the surreal style of the game and presumably intentional.

[1] https://youtu.be/B-Pw72j6xDg

Fun fact: In the Japanese version it's literally just mac TTS voices with machine translated English.

The distortion effects were presumably added so people overseas wouldn't notice how jarring it is. It's such a cool effect it feels like something that should have been in the original release.

Oh wow that’s so interesting! Especially how incomprehensible most of the dialog still is. I just realised this account posted the same video with the non-distorted speech too [1]

[1] https://youtu.be/Jz7FUYRQqfY

That is assuming they tried to write their own TTS instead of just taking an existing, working English TTS and speed it up and distort it. Why would they do that?

To me it does sound exactly like that too in English.

I'd wager it's more obvious in the Japanese version because Japanese is the exceptional language. There are only 44 syllables in Japanese (English has about 16,000) and one would probably still notice this in otherwise unintelligible distorted speech.

Writing their own TTS sounds like what Nintendo would do, to be honest.

Spanish is about as phonetically consistent as Japanese. Most languages have simpler phonology than English. It's not about the number of possible syllables, it's about the rules to go from text to phonemes. The rules in English are immensely complicated and inconsistent. Meanwhile, I can accurately describe Spanish phonology, such that you'd be able to pronounce ~any Spanish word (English loanwords excluded) accurately including stress, in about one page. Written Japanese lacks pitch accent information, but otherwise works similarly.

(By the way, your stats are off; modern Japanese has about ~106 possible syllables (mora) by rough count).

English spelling is inconsistent. English phonology is complex. These are different things.

True, but if they didn't write their own TTS nothing of that matters.

Most TTS software out there will be better at English than any other language despite the more complex phonology. Then it's solely about distinguishing language characteristics and I think the amount of syllables would have an effect in that. Japanese has a lower information density than English (at the same speed) so with the same amount of distortion Japanese should be more recognizable.

Luigi's mansion 3 does something similar, I wouldn't be surprised if it's using the same or similar technology under the hood. For one of the characters, bits of the dialog sounded like incoherent Japanese and I could make out some words. I haven't played in a while, but I recall different characters had different sounding accents too. I like this approach for in-game dialog!

Undertale for some reason manage to do that effect well in English. I don't know how

Reply on news.ycombinator.com