Foreword

When using voice interaction in DeepSeek Harness, relying on external APIs introduces latency and network dependency. Developers need a local recording solution that can run locally and does not upload audio to a server.

What Is This

Parakeet voice input for DeepSeek Harness is a plugin maintained by vTRKA. By integrating Parakeet and NeMo-Speech.cpp models, it enables speech-to-text on a local machine and ensures that audio processing does not leave the local environment.

Core Features

  1. Provides a local microphone button in Composer.
  2. Uses the Parakeet model for offline speech transcription.
  3. Supports shortcut key operation (default Ctrl+E / Command+E).
  4. Supports customizing shortcuts through the settings page.
  5. Audio processing remains on the local machine (127.0.0.1) and is not uploaded to external APIs.

Installation and Enablement

  1. Make sure the system is Windows x64 and that the DeepSeek Harness Web profile is installed.
  2. Open PowerShell and navigate to the plugin directory.
  3. Run the installation script:
.\install.ps1
  1. After installation completes, restart DSH Desktop.

Typical Usage

  1. Recording: Click the microphone button in Composer, or press the shortcut key (default Ctrl+E / Command+E) to start recording. Press the same shortcut key again to stop recording.
  2. Configuring Shortcuts: Right-click the microphone button to select a shortcut, or configure it on the Settings → Plugins → Parakeet Voice Input page.

Applicable Scenarios and Notes

  • Applicable Scenarios: Scenarios that require local speech-to-text and are privacy-sensitive.
  • Notes:
    • The plugin runs with the permissions of the current DSH process. Make sure local port 8080 is not occupied.
    • The browser requires microphone permission. Audio is sent only to the local service 127.0.0.1.
    • During recording, interim words are not displayed. The final transcription result is displayed after the second click.
    • Before installation, please check the source code and license (MIT).

Short Conclusion

This plugin provides fully localized voice input capabilities for DeepSeek Harness, making it suitable for developers who prioritize privacy and offline operation. For more details, see the GitHub repository.