Introduction

DeepSeek Harness Web GUI (dsh web) uses a plugin-based architecture. In the Web interface, voice input and text formatting are important requirements for improving interaction efficiency. This plugin is maintained by LBXC-666 and adds voice input and text refinement features to dsh web, supporting multiple recognition methods and AI-assisted text optimization.

Core Features

The plugin adds a microphone button to the left of the send button in the input box and provides a settings page for configuration.

  1. Voice Input
    - Supports holding the right Alt key to speak; release it and the audio is automatically recognized and inserted into the input box. It also supports clicking the microphone button directly.
    - While recording, the button shows a voice waveform animation, and the recognized text is displayed scrolling above the button.

  2. Recognition Methods
    - Browser local recognition: based on the Web Speech API, free but only supported in Chrome / Edge.
    - OpenAI-compatible API: supports OpenAI Whisper, Groq, SiliconFlow, and other services.
    - Custom JSON API: supports base64 audio transmission.
    - Real-time streaming API: supports transcribing while speaking; requires configuring a local proxy.

  3. AI Text Optimization
    - After recognition is complete, an AI optimization button appears to the left of the microphone.
    - Left-click to run one round of error correction; right-click to run one round of error correction and a second round of formatting.
    - Supports custom prompts, injecting conversation context, and selecting model thinking levels.
    - The optimized result replaces the original text in place, accompanied by a flowing light scan animation.

Installation and Enabling

Use the official plugin command to install it.

dsh plugin --profile web add github:LBXC-666/dsh-voice-input

After installation, you must restart dsh web to load the plugin. After refreshing the interface, a microphone icon appears to the left of the send button in the input box, and a “Voice Input” option appears in the settings left-hand navigation.

Configuration and Usage

On first use, configure the recognition API on the settings page.

  1. Select API mode: Go to Settings → Voice Input, choose browser local recognition or an OpenAI-compatible API (such as OpenAI Whisper), and enter the corresponding API key.
  2. Voice transcription: Hold down the right Alt key to speak. After releasing it, the text is automatically filled into the input box.
  3. Text refinement: Click the AI optimization button inside the input box. Left-click to correct errors; right-click to format. The optimized result replaces the original text in place.

Notes

  • The plugin does not include a built-in transcription service; you must import a configuration or manually configure an API on first use.
  • Browser local recognition is only supported in Chrome / Edge.
  • The real-time streaming API requires configuring a local proxy and DASHSCOPE_API_KEY.
  • API keys are stored only in the browser localStorage, and proxy logs do not print secrets.

Summary

This plugin provides a complete closed loop from voice input to AI text refinement for dsh web. By configuring different recognition APIs, it can meet the needs ranging from local free recognition to high-performance cloud-based streaming transcription. All configurations and keys are stored locally, ensuring safety and controllability.

GitHub Repository | Community Directory