nihongo.stream speech files

On-device speech recognition files for the Kana Trainer game of nihongo.stream, a static website of games for learning Japanese. The website downloads these files straight into the visitor's browser, which runs them offline. No server or account is involved.

Everything here is an unmodified copy of a public release. The layout is made by scripts/fetch-speech.sh in the nihongo.stream repository.

Path	What	Source	License
vosk/vosk.js	vosk-browser 0.0.8 (Vosk/Kaldi in WebAssembly)	npm vosk-browser	Apache-2.0
vosk/vosk-model-small-ja-0.22.zip	Vosk small Japanese model	alphacephei.com/vosk/models	Apache-2.0
vosk/vosk-model-small-en-us-0.15.zip	Vosk small US English model	alphacephei.com/vosk/models	Apache-2.0
sherpa/1.12.35/*	sherpa-onnx 1.12.35 WebAssembly engine (shared by all models below)	k2-fsa/sherpa-onnx release v1.12.35	Apache-2.0
sherpa/1.12.35/moonshine-tiny-ja/	Moonshine v2 tiny, Japanese, with Silero VAD	same release, bundle vad-asr-moonshine-v2-tiny-ja	Moonshine Community License (non-commercial); Silero VAD: MIT
sherpa/1.12.35/moonshine-base-ja/	Moonshine v2 base, Japanese, with Silero VAD	same release, bundle vad-asr-moonshine-v2-base-ja	Moonshine Community License (non-commercial); Silero VAD: MIT
sherpa/1.12.35/reazonspeech-ja/	ReazonSpeech k2 v2 (Zipformer), Japanese, with Silero VAD	same release, bundle vad-asr-ja-zipformer_reazonspeech	Apache-2.0 (reazon-research/reazonspeech-k2-v2); Silero VAD: MIT
sherpa/1.12.35/sense-voice/	SenseVoice small (zh/en/ja/ko/yue), with Silero VAD	same release, bundle vad-asr-zh_en_ja_ko_cantonese-sense_voice_small	see FunAudioLLM/SenseVoice; Silero VAD: MIT

Each sherpa/1.12.35/<model>/ folder holds the bundle's Emscripten loader (sherpa-onnx-wasm-main-vad-asr.js) and its packed model (sherpa-onnx-wasm-main-vad-asr.data). The Moonshine packages contain the full Moonshine license text as /LICENSE; it is also in LICENSE-moonshine.txt here. LICENSE lists the license of every file.