Served now
The server parses with this version, and apps download it from GET /v2/models
(versions newer than it are not offered). A switch takes effect on the next request: no restart.
Installed versions
Compare versions on a phrase
Raw model output of each version, before the server's checks and Gemini. Nothing is saved or switched.
Switch history
Speech engines
Read only: speech models are changed in server/.env + restart (see
setup guide 6). Test them with real audio in the
API console → "Speech only".
ONNX packs (FP16, per language)
Fast; used first for their language. The same files the apps download for offline use
(GET /v2/models, id asr-<lang>). Folder:
Multilingual .nemo model
Which engine each language uses
engine=auto (the default) uses the ONNX pack when the language has one, else .nemo.
Only languages with an ONNX pack work offline in the apps.
Export sources (single-language .nemo)
What the ONNX packs are exported from (scripts/export_asr_packs.py). Not used to
serve requests.