DeepSeek Harness provides pluggable extension capabilities. In the DSH Web interface, interaction can be done not only by keyboard input but also by voice. The dsh-voice-control plugin leverages native browser interfaces to add voice input and speech output functionality to the Web client.
Plugin Introduction¶
The plugin is named dsh-voice-control and is maintained by developer SuCriss. Its core purpose is to turn the DeepSeek Harness Web interface into a desktop Jarvis with voice interaction capabilities. The plugin is licensed under the MIT License.
Core Features¶
The plugin primarily implements the following features:
- Microphone icon: The microphone icon is embedded to the left of the composer send button, with styling consistent with native buttons and no extra space usage.
- Voice input: Supports speak-to-send, or inserting the spoken content into the draft for editing. Based on the Web Speech API, it depends on Chrome or Edge browsers.
- Automatic speech output: Assistant replies are read aloud automatically. The system automatically cleans Markdown formatting for natural speech and triggers playback within 1.5 seconds after the reply is completed.
- Voice compatibility: The voice list is fetched in real time based on the browser and system, grouped by language, and supports fallback across devices.
- Settings and shortcuts: Right-clicking the microphone opens a settings panel (speech rate, voice, automatic send/read-aloud toggles, preview). It also supports the
Ctrl+Mglobal shortcut to start recording.
Installation and Enablement¶
Install it using the official command:
dsh plugin --profile web add dsh-voice-control
After installation, you need to restart dsh web for the configuration to take effect.
Usage¶
- Send voice: Click the microphone icon on the left side of the input box, or press
Ctrl+Mto start recording. After you stop speaking, the content is sent automatically or inserted into the input box. - Adjust settings: Right-click the microphone icon. In the popup menu, you can adjust speech rate, voice, preview, and the toggles for automatic sending/reading aloud.
- Receive voice: After the assistant replies, the content is read aloud automatically. Long replies are truncated to the last 1200 characters for playback.
Applicable Scenarios and Notes¶
This plugin is suitable for users who need to quickly input commands by voice or who prefer voice feedback. The plugin runs entirely in the browser, with zero dependencies, zero backend, and zero API Key required.
Pay attention to the following limitations:
- Browser compatibility: Voice recognition is only supported in Chrome and Edge browsers and is not supported in Firefox.
- Network dependency: Voice recognition requires a network connection (using the browser’s built-in service), while speech output (TTS) is fully local.
- Recording method: Push-to-talk recording is used to avoid picking up TTS audio and creating feedback loops.
- Content length: Automatically read-aloud replies have a length limitation, with a maximum of 1200 characters.
- Multi-session view: It is not adapted for inactive input boxes in side-by-side multi-session views.
Reference Resources¶
Plugin directory page: https://www.skillhub.cn/plugins/SuCriss/dsh-voice-control
Source code address: https://github.com/SuCriss/dsh-voice-control
