Skip to content

webui: add parameter groups for the ASR, VAD, diarization and alignment models - #417

Closed
CryptVenture wants to merge 1 commit into
0xShug0:mainfrom
CryptVenture:pr/params-asr-groups
Closed

webui: add parameter groups for the ASR, VAD, diarization and alignment models#417
CryptVenture wants to merge 1 commit into
0xShug0:mainfrom
CryptVenture:pr/params-asr-groups

Conversation

@CryptVenture

Copy link
Copy Markdown
Contributor

Split out of #374 as requested: missing parameter groups, analysis families. The TTS and music groups, the controls added to existing groups, the dead controls, the default overrides and the range fixes are separate PRs.

The problem

Fourteen ASR, VAD, diarization and forced-alignment families had no group at all in model_params.json, so their entire option surface was reachable only by hand-writing JSON into the fallback box. That includes silero_vad's threshold and min-speech/min-silence knobs — which are the point of that model — and kroko_asr's beam search and hotword biasing.

The change

Groups added: fun_asr_nano, granite5asr, higgs_audio_stt, hviske_asr, kroko_asr, marblenet_vad, nemotron_asr, parakeet_tdt, qwen3_asr, qwen3_forced_aligner, silero_vad, sortformer_diar, vibevoice_asr, voxtral_realtime — 57 controls in total.

Every control is named after the key its family's session actually reads. The shared audio_chunk_mode and audio_chunk_duration_sec controls are read through src/framework/audio/chunking.cpp:167 and :190, which each of these families calls; the rest resolve inside the family's own sources.

No existing group is touched, so nothing that worked before changes.

Validation

python3 -m json.tool webui/configs/model_params.json   # parses
python3 tools/check_loader_catalog_sync.py
# ok: runtime loaders, model_specs, and model_manager_v2 are in sync

Every one of the 57 control names was checked against its family's sources — either a literal "<name>" in the family directory, or the shared chunking helper above.

Scope

One data file, 85 added lines, no deletions. No code change. The generated bundle is deliberately excluded — catalog.ts inlines this file at frontend build time and the bundle is not byte-reproducible, so regenerating it in each PR of this split would make the PRs conflict; happy to send one bundle-regeneration PR once the series lands.

…nt models

Fourteen analysis families had no group in model_params.json at all, so their
entire option surface was reachable only by hand-writing JSON in the fallback
box -- including silero_vad's threshold and min-speech/min-silence knobs,
which are the point of that model, and kroko_asr's beam search and hotword
biasing.

Groups added: fun_asr_nano, granite5asr, higgs_audio_stt, hviske_asr,
kroko_asr, marblenet_vad, nemotron_asr, parakeet_tdt, qwen3_asr,
qwen3_forced_aligner, silero_vad, sortformer_diar, vibevoice_asr,
voxtral_realtime. 57 controls, each named after the key its family's session
reads; the shared audio_chunk_mode and audio_chunk_duration_sec controls are
read through framework/audio/chunking.cpp:167 and :190, which every one of
these families calls.

No existing group is touched.

Validation:
  python3 -m json.tool webui/configs/model_params.json
  python3 tools/check_loader_catalog_sync.py     # in sync
  Every control name checked against the family's own sources.
@CryptVenture

Copy link
Copy Markdown
Contributor Author

Closing for now to stay inside the 3-concurrent-PR policy (see the discussion on #422). Nothing is wrong with the change and CI is green on it; I will reopen it as review slots free.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant