| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| Miteiru_8.2.0_x64-setup.exe | 2026-11-09 | 63.6 MB | |
| Miteiru_8.2.0_amd64.AppImage | 2026-11-09 | 169.8 MB | |
| Miteiru_8.2.0_amd64.deb | 2026-11-09 | 23.5 MB | |
| Miteiru_8.2.0_aarch64.dmg | 2026-11-09 | 33.1 MB | |
| Miteiru-language-cantonese.mtrpack | 2026-11-09 | 8.5 MB | |
| Miteiru-language-hanzi.mtrpack | 2026-11-09 | 11.5 MB | |
| Miteiru-language-japanese.mtrpack | 2026-11-09 | 31.7 MB | |
| Miteiru-language-mandarin.mtrpack | 2026-11-09 | 13.3 MB | |
| Miteiru-language-vietnamese.mtrpack | 2026-11-09 | 615.7 kB | |
| README.md | 2026-11-09 | 3.5 kB | |
| v8.2.0 source code.tar.gz | 2026-11-09 | 2.5 MB | |
| v8.2.0 source code.zip | 2026-11-09 | 2.8 MB | |
| Totals: 12 Items | 360.9 MB | 5 | |
What's New in Miteiru 8.2.0
Changes since v8.1.0.
Miteiru's own word splitting
Japanese, Mandarin and Cantonese subtitles are now split into words by Miteiru's own tokenizer, written in Rust, instead of kuromoji.js and jieba.
- More words come out right. We wrote 1,200 subtitle-style test lines, which two AI annotators split into words independently (disagreements were settled line by line). On the half that was kept back from tuning, the share of lines split exactly right went up:
- Japanese: 48% → 75%.
- Mandarin: 34% → 47% (simplified), 25% → 41% (traditional).
- Cantonese: 35% → 58%.
- Japanese endings stay with their word. 食べさせていただきます is one word, looked up as 食べる, and よ, ね, か and けど are separate words you can click. Labels like
Wi-Fi,USB-C,1,980,23:59and8.1.0stay whole. - Faster and lighter.
- The first Japanese subtitle appears about a quarter of a second sooner, and Japanese lines are split 2.5× faster.
- The background worker holds half the memory with Japanese loaded.
- Mandarin's word list takes a third less memory.
- Your dictionary takes part. Words you add, or that AI word splitting saves, now take part in splitting the line, and the words around them are no longer cut apart. A word you typed in wins over one AI saved. Adding or removing a word changes the subtitle at once.
- Going back. If Japanese splitting looks wrong to you, the old kuromoji.js stays available for this release: set
tokenizer.japanese.useKuromojitotrueand restart. Please tell us what went wrong.
Mandarin study
Two new options in Video settings → Subtitles, for Mandarin. Both are off until you turn them on.
- Tone colours.
- Each syllable takes the colour of the tone it's spoken with. 一 and 不 change with the next word, and so do two third tones in a row.
- Colour the characters, the pinyin or both.
- Two palettes: Standard and Colour-blind safe. A key shows mā má mǎ mà ma.
- HSK levels.
- Words are underlined by HSK 3.0 level (1–6, and 7–9 as one band). A dashed line means the word isn't on the list.
- Levels you already know hides the underlines up to your level.
- The word popup shows the word's level.
- The word list is complete-hsk-vocabulary (MIT).
Under the hood
- The app's internal names no longer mention Electron, which Miteiru 8 left behind.
Downloads
- Windows:
Miteiru_8.2.0_x64-setup.exe. Install it over 8.1.0; your words, settings and downloaded languages stay. - macOS (Apple silicon):
Miteiru_8.2.0_aarch64.dmg. Still not notarized: open it anyway. - Linux:
Miteiru_8.2.0_amd64.deband.AppImage, built on Ubuntu 22.04. Video plays through the system's libmpv: the.debinstalls it; for the AppImage, installlibmpv2(libmpv1on Ubuntu 22.04). - Language packs:
Miteiru-language-*.mtrpack. Miteiru downloads these itself; they're here for installing without a connection. - Updating: 8.2 doesn't update itself. New versions appear on the releases page and at miteiru.hocky.id.
Full Changelog: https://github.com/hockyy/miteiru/compare/v8.1.0...v8.2.0