Introduction

The DeepSeek Harness (DSH) ecosystem already has offline speech-to-text solutions based on local models (such as SenseVoice). These types of plugins usually excel in privacy protection, but require downloading models and configuring a transcription pipeline, which places certain demands on network conditions and performance.

The dsh-voice-input-space plugin takes a different technical approach. It does not rely on any local models and is fully based on the browser-native Web Speech API, achieving minimalist voice input by “long-pressing the space bar.” For developers who want fast input while keeping their hands on the keyboard, this is a plug-and-play solution.

Plugin Overview

The plugin is maintained by developer XSakura666 and is licensed under the MIT license. Its core purpose is: in any editable text box, long-press the space bar to start listening, and release it to insert the transcribed text at the cursor position.

Core Features

  • Long-press trigger mechanism: In an input box or any text area, long-press the space bar (for a duration of ≥450ms) to start listening. Release to automatically insert the transcribed text.
  • Instant cancellation: During listening, press the Esc key or click the “Cancel” button on the panel to abort the current input. This will not affect the current cursor position.
  • Chinese input method compatibility: The plugin includes an isComposing protection mechanism to ensure that the space key used when selecting candidates in a Chinese input method is not intercepted. A short tap on space still behaves as a normal space key.
  • Multilingual support: Supports Mandarin, Traditional Chinese, Cantonese, English, Japanese, and Korean. You can switch languages in “Settings → General.”
  • Status feedback: A status bar below the input box displays the currently selected language and listening state in real time. The feature can also be disabled with one click in the settings.

Installation and Activation

Run the following command in a terminal or command line to install the plugin:

dsh plugin --profile web add dsh-voice-input-space

After installation is complete, refresh the browser page. On first use, the browser will request microphone permission. Click “Allow” in the address bar.

Usage Steps

  1. Click the input box and ensure the cursor is focused.
  2. Long-press the space bar for about 0.5 seconds. The panel will appear and start real-time transcription.
  3. Speak and observe the real-time text in the floating panel.
  4. Release the space bar. The text will be inserted at the cursor position.

Notes and Debugging

  • Privacy and dependencies: Speech recognition is handled by the browser’s Web Speech API. Audio data is not saved locally, but a network connection is required to call the platform’s voice services (Chrome/Edge).
  • Permission handling: If microphone permission is not granted, the panel will continue to display an error message until the key is released.
  • Debug hooks: Developers can access the plugin’s state and counters through the global object window.__dshVoiceInput for debugging.

Conclusion

dsh-voice-input-space offers another approach to voice input: it does not focus on offline privacy, but instead provides zero configuration and a minimal interaction flow. If you are looking for a voice input tool that does not require downloading models and works out of the box, you can try this plugin.

Plugin Directory Page
GitHub Source Code