WORKFLOW 06 · ON-DEVICE SPEECH RECOGNITION

Local voice & audio transcription with no cloud fallback

Dictate complex prompts, transcribe system audio streams, or capture meeting notes on your Windows machine. A one-time speech-model download may be required. Once the model is available locally, transcription can run offline. SafeRelay does not use a cloud transcription fallback.

Ctrl+Alt+W Microphone Voice Ctrl+Alt+S System Audio (WASAPI) Ctrl+Alt+M Meeting Notes No Cloud Transcription Fallback
Interactive walkthrough Microphone / System Audio → Local Speech Model → Transcript
LOCAL SPEECH MODEL · ZERO CLOUD FALLBACK
1. Audio Source Mic or WASAPI

Capture Spoken Context

• Ctrl+Alt+W → Microphone Input • Ctrl+Alt+S → System Audio Loopback • Ctrl+Alt+M → Meeting Notes Session

Start and stop capture with the global shortcut for your audio source.

2. Local Speech Model No Cloud Fallback

On-Device Inference

[Local Speech Engine] Audio frames processed locally after one-time model setup. Cloud API calls: 0

Once the speech model is available locally, transcription runs on your machine without sending audio to external servers.

3. Transcript Ready to Use

Local Text Transcript

"Summarize the deployment steps for the staging cluster and list the rollback checks..."

Use the transcript directly or relay it through SafeRelay's local DLP engine to armed destinations.

THREE LOCAL AUDIO SHORTCUTS

Choose the right audio capture mode

SafeRelay Desktop includes three dedicated shortcuts for voice dictation, desktop audio loopback, and structured meeting capture.

Ctrl+Alt+W

Microphone Voice Transcription

Speak directly into your microphone to dictate prompts, notes, or troubleshooting summaries without typing.

Ctrl+Alt+S

System Audio (WASAPI Loopback)

Capture what is playing through your Windows audio output (webinars, technical briefings, or recorded videos) directly via Windows WASAPI loopback.

Ctrl+Alt+M

Meeting Notes Transcription

Capture longer-form spoken discussions and generate local transcripts for post-meeting review and sanitization.

How Model Setup & Offline Operation Work

SafeRelay is transparent about when network access is and isn't needed for voice transcription:

  • One-Time Speech-Model Download: A one-time speech-model download may be required when initializing local voice recognition for the first time.
  • Offline Transcription After Setup: Once the model is available locally on your machine, transcription can run offline.
  • No Cloud Transcription Fallback: SafeRelay does not silently upload your audio to a remote speech-to-text API if local recognition is busy or offline.

Combine Voice Transcription with Protected Relay

Once your spoken words or system audio are transcribed into text on your machine, you remain in complete control of where that text goes:

  • Review the transcript locally to verify technical terms, names, or numbers.
  • Use SafeRelay's local DLP and Custom Protected Terms (Ctrl+Alt+G) before relaying meeting notes or dictated prompts to armed browser AI tabs or desktop editors.
RELATED WORKFLOWS & DOCS

Explore more SafeRelay workflows

Try local voice & system audio transcription on Windows

Included in SafeRelay Desktop for Windows (7-day full Pro trial included).