(Trying to make 2022 work 
)
Any suggestions for up-to-date chinese (standard) ressources?
My old HSK ressources are so outdated they should be in a museum. The ones I find are either too “street language” for my purposes or they are re-illustrated editions of old-fashioned books with bad structure.
what do you mean by - too “street language”?
I mean it’s either too colloquial (useless in professional settings) or too informal (too much slang, difficult to use in many settings, difficult to keep up with) or full of curse words (improbable for me to use in most settings).
I like learning the current everyday language and slang variations (and curses) but it must be done in a natural way.
Japanese has limited number of possible syllables . Syllables inside words don’t unite or mutate with each other in crazy ways. They just pronounced one after another. So its easy to make Japanese speech robot. While for other languages neural nets are needed.
The tone, length, focus and ease-in/ease-out of sounds and syllables is not interchangeable in each and every instance of speech.
Pre-recording messages may work in specific situations. We say “turn left”, “turn right”, “press nine” using the same tone of voice. The software can do it successfully. If we take the same software and make it say “closed on Sundays” or “how many tickets?”, the result will be awkward.
Except they do.
Wiki has a generic overview of all the exceptions japanese has: Japanese phonology - Wikipedia
Here’s a fun video on the ん sound alone: The 7 different ways to pronounce ん / 「ん」の7つの音 - YouTube
It’s a preview video, but his full video explaining all the rules that determine which pronunciation is to use is 20 minutes long.
Another common offender is が which is often getting nasalised. Another one I can think of is ん nasalising the preceding vowel in some specific cases. And a bunch of other exceptions that, without you knowing the sound shifts that happened from early and old japanese, would never make sense.
The do mutate, but not in crazy ways, and they do it pretty consistently. Even including the 7 ways to pronounce ん, it doesn’t help much if you realise that a language like Dutch has syllables like schreeuwt /sxreːu̯t/ which combine 6 components into a single sound.
Thus, when looking at Germanic or Slavic languages, for example, you’ll find about an order of magnitude more sounds than when you look at Japanese.
Over here there’s a comparison of possible number of syllables:
Japanese: 643
Korean: 1104
Mandarin: 1274
Cantonese: 1298
Basque: 2082
Thai: 2438
Italian: 2729
Spanish: 2778
French: 2949
Turkish: 3260
Catalan: 3600
Serbian: 3831
Finnish: 3844
Hungarian: 4325
German: 5100
Vietnamese: 5156
English: 6949
So English has more than 10 times as many sounds in a syllable.
The issue with this syllable count is that it doesn’t account for consistency, and consistency comes from the phonetic rules of the language. For example, polish would land somewhere close to serbian on this list, but its phonetic rules are so standardised it’s not even comparable to most other languages. Exceptions exist, yes, but when those exceptions can be weeded out through letters alone, it’s much easier to code, than when the exception arises due to some abstract reasons. To make a more precise example, 揚げ (age) has the [g] sound, meanwhile 禿げ (hage) has the [ɣ] sound. Another example 日 (ni) as in 日本 (nihon) has [ɲi] but 二 (ni) as in 二重 (nijyuu) has [nʲi]. Something like this would be very difficult to code because the syllables are nearly the same, and yet they’re pronounced different, just because.
Many more examples can be made, but one thing is certain is that japanese phonology only looks simple and elegant until you start actually learning how to speak the language, which is when you realise that it’s a topic of study as big as pretty much everything else in the language.
This is contrary to my experience. Pronunciation of English still trips me up at times, e.g. when I put the emphasis on the wrong syllable and the word can’t be understood, or where something is suddenly pronounced like it would be in French or something.
I have not once had an issue with Japanese speakers misunderstanding a word I said because I pronounced it incorrectly.
As far as I know, this is both dialect specific and always using [g] is not wrong in any dialect.
It’s like the ending “g” in German words such as fertig, which can be both /ɡ/ or /ç/ depending on dialect, and I’ve witnessed people using both variations pretty much in the same sentence.
But this is regular, in the second case the ni is palatalised in preparation for the jyuu, but not in the first, because hon is not palatalised. I’m willing to bet that people will do this automatically, but of course that’s irrelevant when talking about programming speech synthesis.
Japanese might not be as consistent as Polish (I don’t know any Polish, so I can’t say much about it), but its pronunciation is a lot more consistent than many other languages.
Some of these things are indeed dialect-specific, but the tricky part about dialects is that, while it creates a variety in “acceptable” pronunciation, whenever you break the constancy of pronunciation within that dialect, it immediately sounds off. Imagine a british person going off about “bo’el of wo’er” but then saying “better” as “bedder” (not exactly “d” but whatever, the american version). It is acceptable in the scope of the english language, but it immediately breaks the cohesiveness of the dialect that the person speaks, and it immediately sounds off.
Likewise if your vocaloid speaks in standard japanese and then makes a few of those slipups from a different dialect, it sounds immediately off. From my personal observation, japanese can tell a pitch going up or down just 1 mora too soon or too early comparing to their own dialect and they would notice it immediately, let alone a bigger offence that is mispronouncing a consonant.
Some of them, as you said, make sense physiologically. For example, 漫画 (manga) makes sense as [maŋŋa] once you know that the “n” is nasal. But how do you know that? If you don’t, then [manga] makes more sense to you and it’s something you’d do automatically. In this particular case you could classify it as a rule - n before g is nasal and the g* syllable geminates it. But there are cases where you have no way to tell which “flow” to go with in terms of pronunciation.
For example that 日本, the switch from a dental to a glottal is fairly natural, so there are no indicators as to why it needs to be nasal. There are plenty of examples of this kind of combination in british dialects with the glottal stop, e.g. “nate”, though “n” is alveolar, there isn’t much difference, and the glottal-stopped “a” follows fairly naturally.
Yes, agreed, but I don’t get your claim that it’s as big a problem as it is in every other language, because it just isn’t. Pronunciation of Japanese is comparatively speaking not very complex.
Japanese has fewer consonants than most languages, fewer vowels than most languages, fewer syllables than most languages since these consonants and vowels can only be combined in very few ways, and most of the pronunciation can be determined completely from how it’s written in kana (basically only pitch is missing in the writing).
Compare that to English, where “dough”, “through”, “tough”, “cough”, “plough”, “lough” and “borough” all have distinct pronunciation, to name an example. If I write “pough”, you have no way to know how it’s pronounced. On the other hand, if I only pronounced it as rhyming with “cough”, you don’t know how to write it; it could be any combination of a p followed by a / o and ending in f / ff / ph / uf / uff / uph / ugh.
Such a thing does not happen to the same extend in Japanese by a long way. If I pronounce something in Japanese, you can write it down in hiragana without any difficulty. If I write a word in hiragana, the only thing you wouldn’t know for sure is where the pitch accent falls.
So, Japanese both has very few sounds, as well as a writing system (ignoring the kanji) that is quite regular. I find it very hard to believe that learning Japanese pronunciation is as difficult as an average language.
I wish this paper also studied Japanese, but it does show that of the 17 languages that were included, English is one of the most difficult to read and write.
I don’t think I get this…
but looks like most of them rare or doesn’t matter
knowing less than 150 is actually enough to explain and understand almost anything.
I understand what you mean, but as I said, it’s about as much of an issue as in most other languages, and english might be by far the worst language to draw comparisons to. Most other languages are fairly phonemic, as can be seen in your paper as well. English is a major exception here. So this exception aside, I still stand by my words that the japanese phonetics is as convoluted as it is in most other languages, on average. What I wanted to address with this statement in the first place is that people think that japanese is completely phonemic, which is incorrect. Pitch aside of course, but for the purposes of singing pitch doesn’t matter at all, in fact most if not all japanese artists change the pitch in every word so that it matches the melody of the song, making the language sound very funny.
Speaking of which, english used to be fairly phonemic too, with words like “knight” reading [knɪxt] and all that.
Mostly finished playing Natsuno-kanata. What an enjoyable game.
There’s kind of preface short story called 「日陰に咲く花」 to the game which can be read here or there. It’s mostly fluff. But I like scene in the shed. Because sometimes one also finds some relic of the past and wonders, who were these people, what did they do and how they felt. And how it all now is completely lost to history, forgotten and irrelevant, as if it never mattered.
Nice read while listening the soundtrack. Although literature is hard. Instead of a summer breeze, it felt like pushing a boulder up a hill.
English may be an outlier, but I personally expect that Japanese is an outlier in the other direction. But it’s hard to find evidence for that.
Many places where orthographic depth is mentioned, Japanese hiragana are presented as almost perfectly shallow. Shallower than many of the European languages (esp English and French, but I’ve also seen Greek, German and Italian being mentioned as being less shallow). On the other hand, Japanese is also being presented as one of the most opaque, due to the kanji. Finally, I’ve seen it being claimed that a comparison cannot be made that easily, due to the Japanese script being a syllabary and not an alphabet, and thus measuring orthographic depth may be biased because of this.
So I guess we should perhaps agree to disagree until someone has found a study corroborating that Japanese hiragana are more / less opaque than an average script. 
Well, Greek voice instructions sound like “heefiuWO ntoor portANAW tadgprEsnA eenn” although it is very clear where words should begin and end and what syllable is stressed or accented or merged to its neighbour.
I’m not talking about subtle dialect phonetics or obscure linguistic variations at the post-doctorate level. I’m talking about where the average person would probably press “space” and " ’ " if they were typing the above message.
Robots still need work. AI is not working well. Neural networks are either too slow or they rely on sub-standard feedback. I don’t know how these AI things work, but they look like they are putting effort in the wrong direction. Or they are pre-selling very premature products.
if bots still need work somewhere, therefore Deepmind or OpenAI just didn’t try it yet.
Quality input, quality processing, quality output.
I’m not sure text-to-speech tools get this kind of funding.
