Local Whisper transcription crashes immediately on startup (0xC0000409, all models, all CPU backends)

[TR-Ecotime]

Bug: Local Whisper transcription crashes immediately on startup (all models, all CPU backends)

Environment

• OpenWhispr version: 1.7.5 (Windows installer)
• • OS: Windows 11 Pro
• • • CPU: Intel Core Ultra 7 256V (Lunar Lake), 8 cores
• • • • RAM: 16 GB
• • • • • GPU: Intel Arc 140V integrated (8GB)
          Steps to reproduce

1. Download any Whisper model (tested: tiny 77MB, medium 1.4GB) via Speech-to-Text settings, Local mode
2. 2. Attempt a dictation
      Error shown in-app
      "Local Whisper failed: Error invoking remote method 'transcribe-local-whisper': Error: whisper-server process died during startup: load_backend: loaded RPC backend from [...]/ggml-rpc.dll load_backend: loaded CPU backend from [...]/ggml-cpu-alderlake.dll"

Confirmed via Windows Event Log (Application Error)

• Faulting application: whisper-server-win32-x64.exe
• • Faulting module: ucrtbase.dll
• • • Exception code: 0xC0000409 (stack buffer overrun / fail-fast)
• • • • Crash occurs immediately after "loading model from" log line, before any transcription output
        What I ruled out
• Not model-size dependent: crashes identically with tiny (77MB) and medium (1.4GB) models
• • Not CPU backend dependent: crashes identically with ggml-cpu-alderlake.dll and ggml-cpu-haswell.dll (manually swapped)
• • • Not GPU-related: crashes with --disable-gpu flag too
• • • • Not antivirus: no Windows Defender detections/quarantine events found
• • • • • Not path-encoding related: the model file path (containing a non-ASCII character, "é" in the Windows username) resolves and loads correctly per the log output; crash happens after that
          Expected behavior: whisper-server starts and transcribes locally, as it does for other users on similar hardware.

Current workaround: using Cloud Providers (OpenAI) for both Speech-to-Text and Language Models instead of Local mode.