Introduction

The plugin approach of DSH allows the Web GUI to acquire additional capabilities. dsh-voice-kit is a voice plugin for the DeepSeek Harness Web GUI, primarily addressing two specific issues: recording voice via microphone during the input phase and reading out the assistant’s response during the reading phase.

What is this

  • Plugin name: dsh-voice-kit
  • Maintainer: aaaadrop
  • Positioning: Voice input and read-aloud for the DeepSeek Harness web GUI
  • License: MIT

Core Features

Voice Input

Use the microphone button in the Composer / Input area. The documentation indicates it uses host-side transcription; the transcription results are appended to the draft rather than overwriting existing content. Press Esc to cancel.

Read Aloud

Read-aloud supports three categories of engines:

  1. System voices
  2. Microsoft neural voices
  3. Cloud neural voices

Cloud neural voices support SiliconFlow CosyVoice2, as well as arbitrary OpenAI-compatible audio/speech. Markdown and emojis are removed before playback; long replies are chunked and streamed based on sentence boundaries.

Settings

Provided in Settings → Voice → Read aloud:

  • Voice engine
  • one-tap cloud preset
  • voice
  • rate
  • pitch
  • recognition language
  • engine reachability probe
  • key status

Key status only displays saved or not set and does not show the key value.

Installation and Activation

The installation command provided in the documentation:

dsh plugin add dsh-voice-kit

Steps to activate:

  1. Install the plugin.
  2. Restart DSH Desktop.
  3. Refresh the Web GUI.

Documentation materials also mention:

  • Version 0.4.0 is described as published to npm
  • pnpm typecheck passed
  • 239 unit tests passed

PeerDependencies at least include:

  • @deepseek-ai/cordis ^4.0.1
  • Multiple @deepseek-ai/dsh-* packages, version ^0.1.0-rc.6
  • React ^18.2.0

Typical Usage

Test Read Aloud First

Open:

Settings → Voice → Read aloud

Select an engine first, then press Play test:

  • System voice: The documentation says it always works and is suitable as an offline option.
  • Microsoft neural: Requires proxy / Edge reachability.
  • Cloud neural: Requires a key; you can choose SiliconFlow CosyVoice2 or other OpenAI-compatible audio/speech.

Configure Voice Input

Configure voice input first. Cloud TTS and recognition share the same credential; if voice input is already configured, cloud reading can reuse the same credential.

Record with Microphone

Open the microphone button in the Composer and start speaking. Press the button again or press Esc to stop. The transcribed text will be appended to the draft.

Suitable Scenarios and Notes

Suitable for:

  • Reducing keyboard input in the DSH Web GUI
  • Listening to long assistant responses
  • Configuring voice input and reading simultaneously

Notes:

  • Engine requirements differ: System voices always work; Microsoft neural requires proxy / Edge reachability; Cloud neural requires a key.
  • Key status only shows saved / not set.
  • The plugin runs with the current DSH process permissions; it is recommended to check the source code, dependencies, and license before installing.
  • The plugin itself is under the MIT license; the built-in shared/tsdown.client.ts is adapted from official DSH build tooling (MIT) and the dsh-web-ui libExternal option (Apache-2.0).
  • The community directory is an independent site with no official affiliation with DeepSeek / Huansuan, and should not be considered an official app store.

Links

  • GitHub: https://github.com/aaaadrop/dsh-voice-kit
  • Directory page from plugin clues: https://www.skillhub.cn/plugins/aaaadrop/dsh-voice-kit (Not independently verified in the current round of verified materials; please confirm yourself before visiting)