Introduction: A Silent Surprise in VS Code
Every month, VS Code surprises us with a feature that simplifies our daily life as developers. Recently, the Remote SSH extension was replaced by an even more astonishing capability: native speech recognition integrated into the editor. This discovery reveals an embedded AI model that can transcribe your voice without ever sending data to the cloud.
The Heart of the System – Nemotron 3.5 ASR Streaming 0.6B
When you enable dictation for the first time, VS Code silently downloads a lightweight model called Nemotron 3.5 ASR Streaming 0.6B. This neural network, optimized by Microsoft Foundry Local, runs entirely on your machine, ensuring every spoken word stays local.
Simplified Technical Architecture
The model is built on a compact Transformer architecture tailored to memory and compute constraints. It processes audio streams in real time thanks to a streaming pipeline that eliminates any perceptible latency for the user.
Privacy Advantages: Your Voice Stays Home
Unlike cloud solutions such as Wispr Flow or FreeFlow, no voice data is sent to remote servers. You have full control over what is recorded and transcribed, eliminating risks related to confidentiality.
Comparison with Cloud Services
Cloud‑based tools often require accessibility permissions to monitor your screen. In contrast, VS Code only needs audio access, significantly reducing the exposure surface for sensitive data.
Quick & Easy Setup
To enable dictation, simply launch the “Voice: Enable Dictation” command from the command palette. VS Code automatically handles model download and audio initialization.
- Enable microphone permissions on your operating system.
- Launch the dictation command from the palette.
- Start speaking – transcription appears instantly in the editor.
Integration with Existing Workflows
The native dictation seamlessly integrates with keyboard shortcuts and snippets. You can even combine voice transcription with auto‑formatting extensions for clean code right from the start.
“Local dictation is a real asset for developers who care about privacy without sacrificing productivity.” – Cybersecurity Expert
Limitations & Future Outlook
While Nemotron 3.5 ASR offers high accuracy, it remains less effective on highly varied accents or noisy environments. Microsoft is currently working to optimize the model and broaden its linguistic coverage.
Conclusion: Try Native Dictation Today
VS Code has taken a decisive step by offering local speech recognition that combines privacy and performance. Test this feature in your next project, share feedback, and help improve this promising tool.
