Free alpha · Windows · Local-first · Azure optional

Speech to text for teams that need clear data boundaries.

Skald is a local-first speech-to-text and transcription app for Windows. By default, audio is processed on the device. When cloud processing is needed, Skald can use Azure AI Foundry through your organization’s own endpoint.

  • No telemetry
  • Optional: your Azure endpoint
Skald home screen with recording status and transcript preview

The problem

Speech to text is useful. Data processing must be clear.

Speech-to-text tools can be highly useful, but organizations need clear answers before audio and transcripts enter a workflow:

  • Where are audio and transcripts processed?
  • Which service processes the data?
  • Which region and deployment are used?
  • Is telemetry collected?
  • How does the setup fit existing Microsoft and Azure governance?

Skald keeps the data path explicit: local processing by default, Azure optional, and processing modes visible before use.

Data boundary

Process locally. Extend deliberately.

Skald separates local transcription, cloud transcription, and optional text polishing into clear processing modes. Cloud processing is only used when configured through your organization’s Azure AI Foundry endpoint.

01

Local Transcription

Audio is transcribed locally with Whisper-based models. This is the default mode for stronger data control.

02

Azure Cloud Transcription

Audio is sent to your Azure AI Foundry transcription endpoint. Region, deployment, and access controls follow your Azure setup.

03

Local + Azure Polish

Audio stays local. Only the transcript text is optionally sent to your Azure endpoint for conservative cleanup.

04

Azure + Azure Polish

Cloud transcription and text polishing both run through your Azure endpoint. Intended for scenarios where Azure processing is explicitly approved.

Product

Built for daily speech-to-text workflows.

The interface is deliberately practical: record, review, save, and continue working.

Skald settings with processing modes
Processing modes make data flows visible before use.
Skald transcription settings
Configuration for local processing and optional Azure-based processing.
Skald model management
Local model management for different device capabilities.
Skald recording overlay
Tray and overlay feedback during active recording.

Current alpha

What Skald can do today

Speech to text & transcription

  • Local Whisper-based speech to text
  • Push-to-talk and toggle recording
  • Local audio file transcription
  • Tray app for quick access
  • Local transcript history
  • Configurable output folder

Optional Azure processing

  • Azure Cloud Transcription
  • Local Transcription + Azure Polish
  • Azure Cloud Transcription + Azure Polish
  • Bring-your-own Azure AI Foundry endpoint
  • Conservative polishing designed to preserve meaning

Operations & diagnostics

  • Local model management
  • Portable ZIP build for alpha evaluation
  • Manual diagnostic export for polishing review
  • No telemetry
  • Debug exports exclude audio, credentials, auth headers, endpoints, and full prompts

Privacy principles

Clear control before automation.

  1. Local first. Speech processing can happen locally by default.
  2. Azure optional through your endpoint. Organizations use their own Azure AI Foundry endpoint when cloud processing is needed.
  3. No telemetry. The alpha does not send telemetry.
  4. Data-minimized diagnostics. Debug exports are manual and intentionally limited.

Current status

Free alpha for evaluation and pilots.

Skald is currently available as a free alpha. Core functionality is usable today, while distribution, installer experience, and enterprise operations are still being hardened.

Suitable for

  • evaluation
  • technical review
  • pilot groups
  • feedback from IT, privacy, and business teams

Not yet intended as

  • broad enterprise standard rollout
  • formally certified compliance solution
  • regulated production deployment without internal review
Download coming soon

Roadmap

Harden distribution first. Extend deliberately after that.

Near term

  • Code signing
  • Installer and clean desktop distribution
  • Stability and supportability improvements
  • Settings and hotkey polish
  • Performance guidance for weaker laptops
  • Privacy-controlled support bundle

Next

  • User dictionary MVP
  • Admin templates
  • Configuration export/import
  • Enterprise defaults
  • About, imprint, privacy policy, and third-party notices
  • Optional MSIX / Store distribution

Later possibilities

  • Translation as a separate feature area
  • Additional Azure-based text operations with explicit processing boundaries
  • Team or enterprise deployment if pilot demand justifies it

FAQ

Short answers for evaluation.

Does Skald process speech locally?

Yes. Skald supports local Whisper-based transcription, and local processing is the default mode.

Can Skald use Azure?

Yes. Optionally, Skald can use an Azure AI Foundry endpoint provided by your organization.

Does Skald send telemetry?

No. The alpha does not send telemetry.

What does Polish do?

Polish performs conservative transcript cleanup for readability. It is not intended for creative rewriting or meaning changes.

Is Skald production-ready?

Not yet for broad enterprise rollout. Skald is currently a free alpha for evaluation, testing, and pilots.