Speech models for Lumen
On-device subtitles for Lumen, a local-first Android gallery: tap CC on a video and its speech is transcribed on the phone. Lumen downloads these files on request from its Settings; nothing leaves the phone.
| File | What | Source |
|---|---|---|
whistle.cact |
Whistle, speech to text with word timings (English, German, French, Spanish, Italian, Dutch, Polish), for Cactus's needle engine | Cactus-Compute/whistle, commit 741118687e1f008f06d1351a402a731274789790, unchanged. Apache 2.0 |
vad.onnx |
Silero VAD v6.2.3, finds where people speak (16 kHz), so music isn't transcribed into invented lines | snakers4/silero-vad tag v6.2.3, silero_vad_16k_op15.onnx, unchanged, renamed. MIT |
Both are unchanged copies, hosted together so Lumen downloads from one pinned place.
SHA256SUMS lists each file's checksum; Lumen verifies them after download.
License
whistle.cact: Apache License 2.0 (LICENSE). vad.onnx: MIT (LICENSE-silero-vad). See NOTICE.