WhisperDesk Wiki

System requirements

The app is self-contained (.NET 8 is bundled) — no runtime needs to be installed. Keep GPU drivers current for the Vulkan backend.

Silent deployment

The MSI is per-machine (requires administrator rights), installs to C:\Program Files\DarkCipher Systems\WhisperDesk (since 1.4.2; earlier versions used C:\Program Files\WhisperDesk and are moved automatically on upgrade) and adds Start-menu shortcuts. It works with Intune, ConfigMgr, PDQ, GPO software installation or any tool that runs msiexec.

# Basic silent install
msiexec /i WhisperDesk-1.5.0-x64.msi /qn

# Custom folder, install FFmpeg via winget, skip Ollama
msiexec /i WhisperDesk-1.5.0-x64.msi /qn INSTALLFOLDER="D:\Apps\WhisperDesk" INSTALLFFMPEG=1 INSTALLOLLAMA=0

# Pre-set default library and models folders
msiexec /i WhisperDesk-1.5.0-x64.msi /qn LIBRARYPATH="D:\Recordings" MODELSPATH="D:\Models"

# Verbose log for troubleshooting a deployment
msiexec /i WhisperDesk-1.5.0-x64.msi /qn /l*v "%TEMP%\WhisperDesk-install.log"

# Uninstall — keeps all user data
msiexec /x WhisperDesk-1.5.0-x64.msi /qn

# Uninstall and remove settings and models for the user running it
msiexec /x WhisperDesk-1.5.0-x64.msi /qn REMOVE_SETTINGS=1 REMOVE_MODELS=1

Upgrades

• A newer MSI upgrades in place (same UpgradeCode); settings, models, voices and the library are kept.
• An older MSI is blocked from downgrading.
• Re-running the same version offers Repair or Remove.

Detection rule

For Intune/ConfigMgr, detect by MSI product (UpgradeCode {3fc82978-8263-42a1-9d62-8b858d3c0a9b}), or by registry: HKLM\Software\darkcipher0x\WhisperDesk → InstallDir, combined with the file version of WhisperDesk.exe.

Installer properties

Portable edition

Unzip WhisperDesk-1.5.0-x64-portable.zip anywhere and run WhisperDesk.exe. Keep all files together — the runtimes folder holds the GPU engine. No install, no admin rights. User data still goes to %LOCALAPPDATA%\WhisperDesk.

FFmpeg, models & Ollama

FFmpeg

Used to read audio/video and to burn subtitles. WhisperDesk looks for it on PATH, next to WhisperDesk.exe, or at the path in File → Settings → Tools & storage. FFmpeg is not bundled.

Whisper & speaker models

On first use WhisperDesk downloads the selected Whisper model (Large-v3 Turbo ≈ 1.6 GB) and the speaker models (≈ 32 MB) into the models folder. For offline machines, pre-stage the models folder (copy it from a machine that has downloaded them) and point MODELSPATH at it.

Ollama (optional)

Local AI summaries and consolidated reports use an Ollama server — default http://localhost:11434. If you point it at another host, transcripts are sent there; every summary records where it was processed.

File locations

Help → Privacy & security inside the app shows the actual resolved paths for the current user.

Settings reference

Stored per user in settings.json; edited in File → Settings (Ctrl+,).

Network & firewall

WhisperDesk never phones home. Outbound traffic happens only for features a user triggers:

Inbound: the optional caption server add-on listens on localhost:7890 only (off by default). No inbound firewall rule is needed or created.

Privacy & compliance

• Local by design. Audio, transcripts, voiceprints, summaries and settings stay on the PC. No account, cloud service, telemetry, analytics or crash reporting.
• Voiceprints are biometric data (e.g. GDPR Article 9, Illinois BIPA). Obtain consent, retain only what you need, and delete with Tools → Voice library → Forget. Voiceprints are never included in shared files or reports.
• Recording consent law varies (one-party vs all-party). Make sure users tell people when they record.
• Cloud-synced folders: if the library is inside OneDrive, Dropbox, Google Drive or iCloud, that service uploads it. WhisperDesk warns about this; relocate or exclude the library where recordings must remain local.
• At rest: files are protected by the Windows account ACLs. Use BitLocker on the drive and password-protected backups for off-device copies.
• Integrity: text edits are audited (who, when, why, before/after); reports and analysis exports carry SHA-256 fingerprints with a manifest.sha256.txt; backups are verified on restore.
• Add-ons are .NET code running with the user's permissions. Consider controlling %LOCALAPPDATA%\WhisperDesk\addons with App Control/AppLocker if users shouldn't add their own.

Forensics view & audit trail New in 1.5

Open a recording in the player and switch ☰ Transcript / ∿ Forensics, or in the Library right-click → ∿ Forensics….

Tamper-evident audit log

Every action (opened, measured, identified, note, integrity check, export) is appended to forensics-log.jsonl in the recording's library folder. Each entry records the Windows user, time, computer name, the media file's SHA-256 and the hash of the previous entry, so the entries form a chain. Verify chain recomputes it and reports any entry that was edited, deleted or reordered.

• Back up or archive the log together with the original media; the chain proves the log is intact, and the SHA-256 ties it to that exact file.
• The log is plain JSON Lines, readable by any tool. Don't edit it by hand: any change breaks verification (by design).
• Work on a copy of the evidence where possible, and keep the original read-only.

Backup & migration

• File → Backup… writes one file containing the library, voice library, settings and add-ons (models optional).
• With a password it's a .wdbackup encrypted with AES-256-GCM (key from PBKDF2-SHA256, 600,000 iterations); without, a plain .zip.
• File → Restore from backup… merges recordings (existing ones skipped unless overwrite is ticked), merges voices and verifies every file against its SHA-256.
• A forgotten backup password cannot be recovered.

To move a user to a new PC: back up, copy the file across, install WhisperDesk, then restore. Leave models out of the backup to keep it small — they download again.

Identifiers

These never change between versions — safe to use in detection rules, policies and scripts.

Code signing & Smart App Control

Current releases are not code-signed. On managed fleets:

• Smart App Control is disabled automatically on enterprise-managed devices, so it won't block deployment there.
• If you use App Control for Business (WDAC) or AppLocker, allow WhisperDesk by file hash or path (%ProgramFiles%\DarkCipher Systems\WhisperDesk\* (1.4.2+; %ProgramFiles%\WhisperDesk\* for older installs)) — publisher rules aren't possible without a signature. Hash rules must be refreshed for each release.
• Verify downloads against the SHA-256 on the download page, or build from source and sign with your organisation's certificate.

Consumer guidance for Smart App Control is on the downloads page.

Troubleshooting

Dev environment

Get the code from the Source Code page. The add-on contract lives in a sibling project, WhisperDesk.Addons, so keep both folders side by side.

Building

# Debug build and run
cd WhisperDesk
dotnet build
dotnet run

# Full release: MSI + portable zip + guide PDF into .\dist
.\build.ps1

build.ps1 runs dotnet publish -c Release (self-contained, win-x64) into publish\, builds Installer\WhisperDesk.wxs with WiX and the UI extension, zips the publish folder as the portable edition and copies the guide PDF.

Project layout

Architecture

Add-on events are dispatched by AddonManager on a background queue, one at a time and in order; exceptions in add-ons are caught and logged.

Dependencies

Models: OpenAI Whisper (MIT), pyannote segmentation and WeSpeaker embeddings via sherpa-onnx, Silero VAD (MIT). FFmpeg and Ollama are separate, optional programs and are not redistributed.

Release process

1. Bump <Version>, <AssemblyVersion>, <FileVersion> and <InformationalVersion> in WhisperDesk.csproj. The MSI version follows it, so newer MSIs upgrade older installs.
2. Update the "What's new" section in docs\make_guide.py and regenerate the PDF.
3. Run .\build.ps1 → dist\WhisperDesk-<ver>-x64.msi, portable zip and guide.
4. Test an upgrade over the previous version and confirm settings, voices and library survive.
5. Publish the artefacts with their SHA-256 hashes (this website: tools\sync-downloads.ps1).

Add-on SDK

Add-ons are .NET 8 class libraries implementing IWhisperDeskAddon from WhisperDesk.Addons.dll. Full walkthrough, downloads and examples are on the Add-ons page.

public interface IWhisperDeskAddon
{
    string Id { get; }            // unique, stable, e.g. "yourname.word-export"
    string Name { get; }
    string Description { get; }
    string Author { get; }
    IReadOnlyList<AddonOption> Options { get; }
    void Start(IAddonHost host);  // subscribe to host events
    void Stop();                  // unsubscribe and release resources
}

public sealed record AddonOption(string Key, string Label, string DefaultValue, string Help = "");

Reference the contract with Private=false so it isn't copied to your output, and don't ship WhisperDesk.Addons.dll with your add-on.

Caption server API

Provided by the built-in Live captions for OBS & browsers add-on. Listens on http://localhost:7890 (configurable), localhost only.

// Follow the live transcript from any web page or Node script
const es = new EventSource("http://localhost:7890/events");
es.addEventListener("line", e => {
  const line = JSON.parse(e.data);
  console.log(`${line.Speaker ?? "?"}: ${line.Text}`);
});

Webhook payloads

The built-in Webhook add-on sends an HTTP POST with a JSON body to the configured URL (10-second timeout; failures are written to the add-on log).

{
  "type": "line",            // "line" | "saved" | "started" | "stopped"
  "app":  "WhisperDesk",
  "data": {                  // TranscriptLineEvent, TranscriptSavedEvent or SessionEvent
    "sessionId": "…",
    "isLive": true,
    "start": "00:01:12.4000000",
    "end":   "00:01:15.9000000",
    "text": "Welcome back to the stream",
    "speaker": "Luke",
    "sex": "Male",
    "language": "en"
  }
}

Contributing

• WhisperDesk is MIT-licensed — fork it, change it, ship it; keep the copyright and licence notice.
• Match the existing style: small focused services, XML doc comments on public members, no telemetry, nothing leaves the PC by default.
• Bug reports, patches and add-on ideas: the bug report form or the Discord.