[Plugin] mimo-tts.koplugin - continuous TTS read-aloud for KOReader

[joey5403]

mimo-tts.koplugin turns KOReader into an audiobook player: it extracts the book text (current page, a selection, or the whole book), synthesizes speech with Xiaomi's MiMo TTS API, and plays it with the best audio backend available on the device.

Upfront: the text is sent to Xiaomi's cloud API for synthesis - the same trust model as KOReader's built-in translate. The service is currently free during its beta.

GitHub: https://github.com/joey5403/mimo-tts.koplugin
Releases (extract into plugins/): https://github.com/joey5403/mimo-tts.koplugin/releases

What it does

• Read-aloud with a moving underline - one sentence per segment; the sentence being spoken is underlined, and the underline moves on as soon as it has been read.
• Continuous listening - automatic page turns to the end of the book; the next sentences are synthesized while the current one plays, with look-ahead and cross-page prefetching, so playback is effectively gapless.
• Sentence-level cache - audio is cached by content hash, so re-listening a chapter costs zero API requests.
• Three synthesis models - preset voices (Bingtang, Jasmine, Soda, ...), voice design from a text description, and voice cloning from an mp3/wav sample. Style control via audio tags like (gentle) or natural-language style prompts.
• Two independent speeds - a synthesis-side speed instruction written in plain language (any language: 语速稍快, speak a little faster) and a player-side time-stretch.
• Where it plays - on Android it uses an OpenSL ES backend driven through LuaJIT FFI: no external binaries, works on every device (tested on Android 16, arm64). Elsewhere it auto-detects a command line player (mpv, ffplay, mpg123, aplay, ...) - fine on desktop Linux/macOS out of the box; on Kobo/Kindle/PocketBook you need to install a player first (testers for this path are very welcome!).
• Optional streaming synthesis (stream + pcm16), gesture/keyboard bindings via the dispatcher, UI and docs in English and Chinese.

Demo recording (37 s, with audio) - the sentence being spoken is underlined:
watch the recording

Install: grab the release zip, extract the folder into plugins/, add a MiMo API key from mimo.mi.com. Tested on Android 16 (arm64) and desktop Linux.

Feedback and reports are very welcome - especially from Kobo/Kindle users, since I could only test the command-player backends on desktop. Happy to answer questions here.