Introduction

Under the DSH plugin extension model, the assistant responses for dsh web are typically presented as text. The reading effort can be relatively high for long responses or when results need to be checked frequently. dsh-tts is a dsh web text-to-speech (TTS) plugin: it reads out the final result of each assistant response using Edge TTS Microsoft neural voices. It is maintained by vim0x3c, licensed under MIT, and the current version is stated as 0.1.0.

Plugin Positioning

The positioning of dsh-tts is as a dsh web TTS plugin, targeting scenarios where users wish to hear the assistant’s final response and reduce reliance on pure text reading. It does not replace the chat interface nor does it provide voice input.

Core Features

  • Reads out the final result of each assistant response using Edge TTS Microsoft neural voices.
  • Only reads the final result, ignoring intermediate steps and tool invocation descriptions; for continuous multi-turn conversations, only the latest response is read.
  • Long responses are automatically segmented into sentences and read sequentially.
  • Built-in 20+ Edge neural voices, defaulting to zh-CN-XiaoxiaoNeural, supports speed adjustment from 0.5x to 2x.
  • The input toolbar provides a speaker button for quickly toggling “Auto-read new replies”; closing it interrupts the current reading.
  • Supports reading approval notifications, which can be toggled independently, and allows using preset sounds (“Ding/Beeb/Thud”) or uploading custom approval sounds (mp3/wav/ogg, ≤1MB).
  • Falls back to the browser’s local speechSynthesis / Web Speech API when automatic reading fails.
  • The host provides a local bridge service bound to 127.0.0.1, scanning ports 8765-8780, containing /health and /synthesize.
  • The plugin itself contains no third-party npm dependencies and can be installed offline.
  • Browser-side settings are persisted in dsh-tts.settings in localStorage.

Installation and Enabling

Prerequisites: The dsh application closure must include the plugin’s @deepseek-ai/dsh-* peer package; any deployment containing the @deepseek-ai/dsh-web-app bundle satisfies this.

Documentation states the most common method is local tarball installation; using the -w flag is required to install to the web profile. After obtaining dsh-tts-0.1.0.tgz, execute:

dsh plugin --profile web add -w ./dsh-tts-0.1.0.tgz

After installation, you can verify using the following two commands: the first checks if dsh-tts appears in the config, the second checks the host bridge’s /health interface.

dsh --profile web --dump-config | grep dsh-tts
curl http://127.0.0.1:8765/health

The documentation also mentions a git URL installation method, but notes it is not recommended: it retrieves source code rather than the built lib/, and the plugin’s peer packages are not published to npm, making it usually impossible to build or resolve.

Typical Usage

  1. Open “Settings → Text-to-Speech” in dsh web.

  2. Turn on “Auto-read new replies” and click “Preview Voice” to confirm connectivity to Edge.

  3. Subsequently, the response will be automatically read every time the assistant provides a final answer.

Supplementary Operations:

  • Use the speaker button in the input toolbar to toggle “Auto-read new replies”; closing it immediately interrupts the current reading.
  • Approval notification reading can be toggled independently, and preset or uploaded approval sounds can be set.
  • If browser autoplay is blocked, click “Preview Voice” in settings once first.

Suitable Scenarios and Notes

  • Suitable for scenarios where you wish to hear the assistant’s final results in dsh web, process long responses, or handle approval notifications.
  • Edge cloud synthesis requires a network; the host must be able to access speech.platform.bing.com; it falls back to the browser’s local Web Speech API on failure.
  • Only reads the current session; responses for background sessions will not be read.
  • The in-process bridge starts and stops with the dsh web host lifecycle.
  • CORS only reflects dsh web origins within the whitelist; unknown webpage origins return 403.
  • Voice input (microphone/STT) is not included.
  • dsh-tts runs with the permissions of the current dsh process and starts a local bridge service on the host side; it is recommended to check the source code, license, and runtime environment before installation.

Summary

dsh-tts adds a narrow but specific capability to dsh web: reading the assistant’s final results using Edge TTS, while retaining approval notifications, manual toggles, and Web Speech fallback.

  • GitHub: https://github.com/Vim0x3c/dsh-tts
  • Directory Page: Verified documentation does not provide a URL