Introduction

When chatting with models in the DSH web interface, long prompts usually require typing word by word, whereas long replies require continuously watching the screen. For DSH users or AI agent developers, both steps can be made more streamlined.

dsh-chatvoice is a DSH plugin that connects the browser’s native voice capabilities to the dsh web interface: you can speak using a microphone when inputting, and read aloud with one click when receiving replies. Below is an introduction to its positioning, installation methods, typical usage, and limitations to note.

What is this

dsh-chatvoice is maintained by FuzzySoul and is licensed under MIT.

It adds a “voice input + AI reply reading” loop to DeepSeek Harness (dsh). Based on the browser’s native Web Speech API, it requires zero configuration, zero cost, and no API key. The plugin itself has no network requests, no sub-processes, and no API keys.

Core Capabilities

Voice Input

After clicking the microphone button on the input toolbar, the browser will request microphone permission.

While speaking, recognition results enter the input box sentence by sentence, with intermediate results displayed in a real-time bubble above. During listening, you can type at any time to correct typos or delete text; the voice only appends to the end of the input box, and stopped input does not refill deleted content.

Reply Reading

Each assistant reply will have a speaker button next to it.

Click once to start reading, click again to stop, and you can interrupt at any time.

Auto Reading

After enabling it in the settings page, new replies will be read aloud automatically after completion, and can also be interrupted at any time.

Settings Page

Enter ChatVoice in the dsh settings to configure:

  • Recognition Language
  • Auto Reading
  • Voice/Timbre
  • Speed

Changes take effect upon saving; no restart is required.

In Chinese scenarios, the plugin defaults to zh-CN recognition and automatically selects the built-in Xiaoxiao Online (Natural) free Chinese natural voice in Edge.

Error Messages

The following situations will display readable toasts instead of failing silently:

  • Microphone permission denied
  • Browser not supported
  • Insecure context
  • Recognition network failure

Installation and Enablement

Installation command:

dsh plugin --profile web add dsh-chatvoice

Restart dsh web after installation:

dsh web

Open:

http://127.0.0.1:3080

You must use 127.0.0.1 to access here. Voice recognition requires a secure context, i.e., HTTPS or localhost.

If connected directly via LAN IP, the browser will disable the microphone; the plugin will automatically disable input functions and show a prompt, but reading aloud remains available.

Typical Usage

  1. Voice Input: Click the microphone button on the input toolbar, allow browser microphone authorization, and start speaking; click the microphone again to stop, then press Enter to send. During recognition, you can type to correct words or delete, voice only appends, does not refill.
  2. Read a specific reply: Click the speaker button next to the assistant reply to start reading; click again to stop.
  3. Enable Auto Reading: Go to Settings -> ChatVoice, check “Auto Read New Replies”, and it takes effect immediately upon saving.

Applicable Scenarios and Notes

Suitable for scenarios where prompts are frequently input in DSH web, or where you want to turn long replies into read-aloud output. Chinese scenarios are prioritized, and the documentation suggests trying Edge.

Usage Notes:

  • Chrome’s voice recognition goes through Google servers; domestic networks may report network errors; the documentation suggests switching to Edge (which uses Azure).
  • Edge online voices require an internet connection; offline, it falls back to the system local voice.
  • Firefox/Safari do not support SpeechRecognition; the button will automatically be grayed out with a prompt, but reading aloud remains available.
  • Voice recognition accuracy depends on the browser and system microphone, not on the plugin itself.
  • Plugin configuration is persisted to:
~/.dsh/chatvoice.json

This plugin runs with the current dsh process permissions and is used for the dsh web client; it is recommended to check the source code before installation and confirm its MIT license.

Links

  • Directory Page: https://www.skillhub.cn/plugins/FuzzySoul/dsh-chatvoice
  • GitHub: https://github.com/FuzzySoul/dsh-chatvoice