AI Agent Hub
Back to plugins
🤖

dsh-speech-plugin

Model Inference Updated 2026.08.16

Run the following command in DeepSeek Harness:

dsh plugin install huangdejie/dsh-speech-plugin

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install huangdejie/dsh-speech-plugin in your DeepSeek Harness terminal to install the plugin; the repository is available at https://github.com/huangdejie/dsh-speech-plugin .

About this plugin

DeepSeek Harness ships a text-first Web UI. After a long session of reading dense replies, your eyes tire and typing out the next prompt feels slow, especially in meetings or while commuting. dsh-speech-plugin is a standalone add-on that layers voice playback on every assistant message, a per-conversation auto-announce toggle, and a microphone input in the prompt box, all without modifying a single line of harness source code.

Under the hood, playback triggers per message or automatically, strips markdown, emoji, code blocks, and images before speaking. Voice input streams a live transcript into the draft box, auto-punctuates at sentence boundaries, and stops cleanly the moment you tap send, leaving no trailing audio or wasted quota. The TTS engine and the ASR engine are fully decoupled: each can independently target Alibaba DashScope or Volcengine, or both can point at the same provider. When a cloud credential is missing, expired, or a call fails, playback falls back to the browser's system voice and the mic button disables itself with a clear reason, so the experience never hard-crashes.

It is aimed at anyone who lives in the Harness Web UI and reads long replies for hours, or who prefers to speak a prompt instead of typing it. With a single cloud API key and zero config files, the default auto mode just works out of the box. Power users who hold keys from both DashScope and Volcengine get independent engine fields, per-provider model and voice tuning, and a cost gate on text length, giving full control over which provider handles which job.

Use Cases

  • Hands-free listening of long replies during a meeting or commute
  • Speaking a prompt into the mic instead of typing on the go
  • Graceful fallback to the browser system voice when a cloud credential expires

Best For

  • Daily Harness Web users who read long replies for hours
  • Anyone who prefers voice-over-typing in meetings or transit
  • Power users with DashScope or Volcengine keys who want per-scenario engine switching