简体中文 | English

This repository contains practical tools and reverse-engineering scripts for an OLYMPUS DR Series USB/HID device on Windows.

Primary workflow at repository root:

• dictation.py
  • Main dictation runtime (sherpa-onnx + SenseVoice).
  • Hold FAST_BACKWARD to start recording, release to run ASR and type text.
• read_media_keys.py
  • Real-time HID reader and bitmask parser.
  • Source of truth for current button-bit mapping.
• keys_monitor.py
  • Human-readable state transitions for key validation.
• print_current_state.py
  • Raw HID snapshot/debug output.
• create_dictation_shortcut.py
  • One-command setup helper (.venv, dependencies, desktop shortcut).

Runtime actions configured in dictation.py:

• Root directory:
  • Runtime scripts, setup script, key mapping file, and capture files (*.pcap).
• analysis/:
  • Protocol/pcap analysis scripts and replay experiments.
  • qwen3_integration_research.md: quick feasibility research for Qwen3/Qwen3-ASR integration.
  • _tmp_*.py files are scratch probes used during reverse engineering.

• Target device: VID:PID = 07B4:0256.
• Input state comes from HID interrupt IN reports.
• INSERT_OVER transitions correlate with host HID SET_REPORT(Output) on EP0.
• Observed LED-related control pattern:
  • bmRequestType=0x21
  • bRequest=0x09
  • wValue=0x0200
  • wIndex=0x0000
  • wLength=64
• Replaying candidate 64-byte output reports still does not reproduce LED behavior reliably; device/session context is likely missing.

• OS: Windows
• Python: 3.12
• Recommended: local .venv

Dependencies are listed in requirements.txt.

python -m pip install -r .\requirements.txt

Main packages:

• pywinusb
• sherpa-onnx
• sounddevice
• numpy
• pynput

Create environment + install deps + desktop shortcut:

python .\create_dictation_shortcut.py

Monitor HID reports:

python .\read_media_keys.py

Run readable key monitor:

python .\keys_monitor.py

Start dictation runtime:

python .\dictation.py

Download and extract the SenseVoice model (or pass --model-dir to point to your path):

https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2

Supports Chinese, English, Japanese, Korean, and Cantonese with automatic language detection and ITN (inverse text normalization).

• Analysis scripts are intentionally separated under analysis/.
• Scratch scripts can be cleaned up later if you want a stricter production-only repository layout.