Sentences with native audio (esp. Context)

I don’t know if it will be too much work for WaniKani team, but as far as I have seen,

  • Native audio for sentences are rare. Nor are they on Forvo.
  • TTS for sentences in Japanese’s case is a little faulty, like choosing a wrong reading for the Kanji. Nor have I seen that any TTS has Furigana support.
  • There is [Userscript]: Anime Context Sentences, but searching for the vocab in sentences is not humanly made, so not checked for proper results; like sometimes it’s not checked if it is outside the speech or in names.
    • Also, I am not sure if is possible to extract audio. Furthermore, licensing to use elsewhere, for example, Anki or some other Japanese learning websites / apps.
    • Not to mention that the video is sometimes not properly chopped or out-of-sync.

About the amount of context sentences, there are a little less than 3 times of the number of vocabularies here, I guess.

3 Likes

Are you looking for sentences with native audio? Supernative does that I guess if you don’t wanna just watch a show out right

1 Like

I would love to have sentences with native audio as well, but I don’t think the WK team is going to implement it.

Even some of the content updates didn’t have item audio until later and those are much less work than full sentences would be.

2 Likes

Thinking about the resource making. It might be more proper to find a native audio sentence database (with or without video), and try to caption by myself. It is doable just like gradually adding vocabularies to Anki.

About the process, it can be done with Speech Recognition (or a real transcript) in advance, then proofreading during individual vocabulary learning.

Anime context sentences is made with https://immersionkit.com which is an external resource with a database of voiced sentences you can check for yourself.

Tofugu made a few cool things previously like releasing their single vocab audio under Creative Commons GitHub - tofugu/japanese-vocabulary-pronunciation-audio. However we’re yet to hear from them about context sentences.

Actually, bunpro has a lot of voiced sentences that are also split by words and are somewhat synchronized with wanikani vocab. But it’s uncertain how soon will it be released or ported someway to wanikani if ever.

I honor the ambitions but I do not see how it will solve the problems you pointed out in the opening post. It will require the same immense amount of proofreading as the other databases.

Apparently, there is ImmersionKit API (with only one endpoint). However, it isn’t said on the website. I look into that UserScript’s source code instead.

About pulling from YouTube transcript, it is possible too; but actually I won’t do it yet, as it doesn’t seem to worth the effort that much. I will use Microsoft Azure TTS for now.

Actually, as of now, my Anki card is “EN → JP + autoplay sentence audio” and “JP → EN + autoplay vocab audio”, so yeah, I do need sentence with audio, if possible, for every vocabularies.

I also realized about this recently, due to Jotoba - A new alternative to jisho!

Theres plenty of tools out there that take subtitle tracks and break up the audio of an episode for that subtitle track. If you want to make a database of native sentences with audio, that right there would give you context, audio, text, and even visuals if you wanted.

1 Like

The problem really lies in having a big enough database, that also accounts from vocabularies I gathered from non-listening methods; that is, aren’t originally from movie sentence mining.

Furthermore, even if the vocabulary is from listening, it doesn’t account for additional meanings / collocations. More sentences need to be found, anyway.

Nonetheless, I agree that I need to do sentence mining (with audio) at some point.

I talked with the author on Discord, and got the website fixed to what I wanted. So, I don’t really need to use the API.

Actually, I found your post - 🍃 Shenmue Tree - a Study "Lounge!" 🍂 - #307 by Vanilla but I still have to evaluate if I can make an actual searchable database. Apparently, audio is also a must at this point. Japanese sentence searching mechanism I am seeing now is probably PGroonga.

About ImmersionKit, I have grown dislike for cutesy Anime voices and goes for dramas instead. (Something like Death Note is still OK, though.) I also would consider going for news, documentaries and live actions. (Japanese songs, non-Anime ones, are ok as well, but I wonder if I should pursue them.) A part of it is, I don’t like Japanese pop culture that much any more.

I also found the case of bad searching in ImmersionKit, as well as, (almost) not found, so I indeed need to move on.