Russian voice input for DSH — local GigaAM-v3 provider (offline, MIT)

[almaziphone]

Russian speech input for DSH — local, offline, MIT

The built-in Voice input recognizer (SenseVoiceSmall) doesn't cover Russian: it supports
zh / en / yue / ja / ko only, and there is no cloud ASR in the voice-input composition. So I
wrote a small plugin that registers a second local provider on top of the sherpa-onnx
runtime DSH already ships — no cloud, no API keys, no Python.

• Model: GigaAM-v3 e2e CTC INT8 (MIT, SaluteDevices/Sber) — Russian SOTA; the e2e variant emits
  punctuation, casing and normalized numbers («в 15:30»)
• Quality/speed measured on the reference set: 3.0 % WER, RTF 0.017 @2 threads
  (8.4 s phrase → ~0.17 s), ~564 MiB peak RSS
• 100 % local: audio never leaves the machine; the model is a pinned on-demand download
  (321.7 MB + 1.8 MB VAD) verified by SHA-256
• Works with the standard voice UI: recognition cards, preparation progress/cancel, provider and
  language selection in Settings → Voice input; Whisper Small is available as a multilingual
  fallback engine
• MIT, no new native dependencies

Install on DSH Desktop:

bash -c "$(curl -fsSL https://raw.githubusercontent.com/almaziphone/dsh-speech-to-text-ru/main/scripts/install.sh)" install.sh

or manually: enable Plugins → Voice input, then
dsh plugin --profile desktop add github:almaziphone/dsh-speech-to-text-ru, refresh the window
(Cmd+R / Ctrl+R). After the first preparation the recognizer appears as
«GigaAM-v3 · RU (int8)» with language ru.

Repository (public, MIT): https://github.com/almaziphone/dsh-speech-to-text-ru — benchmarks,
per-clip transcripts, model hashes and design notes are in docs/.

Если удобнее по-русски: скриншоты/описание и быстрый старт — в
README.ru.md.
There is also an upstream proposal in
Ideas #11745 if the team ever
wants to ship this as an official optional bundle.

Feedback and PRs on the plugin repo are welcome. — @almaziphone