Repository navigation
Captions pack 3 (whole sentence speech models: Japanese, Filipino, Lithuanian, Auto)
Pre-releaseOptional speech models for captions in Japanese, Filipino, Lithuanian and any other language (Auto) in PurpleTV ReVive (ad break captions and live captions). You do not need to download anything here by hand: pick the streamer's language under PurpleTV › Player › During ad breaks › Stream language (or under PurpleTV › Translation), then tap "Download speech model". The module checks every file against its pinned size and SHA-256 and only then uses it. Everything runs on your phone, and no audio or text is sent anywhere.
The captions engine (sherpa-onnx) and the English model stay in captions-pack-1; you need that engine too.
These models hear a whole sentence at a time, so their captions come a moment after the words are said. Each zip also holds Silero VAD, which finds where each sentence starts and ends.
This is a pre-release on purpose, so it never shows up as the module's latest version.
What is in here
| File | Stream language | Download | On the phone |
|---|---|---|---|
purpletv-captions-model-reazonspeech-k2-v2-ja.zip |
Japanese | 139 MB | 171 MB |
purpletv-captions-model-omnilingual-asr-300m-ctc-int8-2025-11-12.zip |
Filipino, Lithuanian and Auto (one model for more than 1600 languages) | 291 MB | 367 MB |
Each zip holds the model files exactly as published at the sources linked below (nothing changed or retrained), plus a LICENSE.txt and a README.txt with the credits, the source commit and every file's SHA-256, and the full licence texts.
Credits and licences
- ReazonSpeech k2 v2 by Reazon Human Interaction Lab (Japanese), Apache-2.0: https://huggingface.co/reazon-research/reazonspeech-k2-v2
- Omnilingual ASR (omniASR_CTC_300M) by Meta (Filipino, Lithuanian and Auto), Apache-2.0: https://github.com/facebookresearch/omnilingual-asr. Converted for sherpa-onnx by csukuangfj: https://huggingface.co/csukuangfj/sherpa-onnx-omnilingual-asr-1600-languages-300M-ctc-int8-2025-11-12
- Silero VAD by the Silero Team, MIT: https://github.com/snakers4/silero-vad
- sherpa-onnx by k2-fsa, Apache-2.0, runs them all: https://github.com/k2-fsa/sherpa-onnx
Thank you to everyone behind these projects!