Add local LLM voice model transcription#1

Open

pdufour wants to merge 17 commits intomainfrom

paul.dufour/test-local-trans

Commits on Mar 15, 2026

wip: rip out openai-only transcription, wire up transcribe-rs with parakeet onnx backend
pdufour
committed
download missing parakeet model files from huggingface on first use
pdufour
committed
add /voicemodel slash command, voice model popup, bring back openai transcription path
pdufour
committed
restore chat_composer.rs voice keybinding block that got nuked
pdufour
committed
switch model download from curl to hf-hub api, add tracing, kill /tmp debug log
pdufour
committed
clean up voice model validation, replace alias map with allowed-id list, trim chatwidget
pdufour
committed
drop normalize_selected_model and validate wrapper, inline validation into try_set
pdufour
committed
add 24k to 16k wav resampling for local models, simplify voice popup and set_voice_model
pdufour
committed
revert .gitignore additions for model artifacts and sample.ogg
pdufour
committed
restore openai transcribe duration_seconds param and 24k sample rate test expectations from main
pdufour
committed
fix normalize_chatgpt_base_url: use parameter name base_url instead of input
pdufour
committed
voice: repo-to-class map, generic SpeechModel loading, voice_model_picker_options()
pdufour
committed