A DSH Web GUI Sound Lab: pick a sound for each of the four triggers (session end / options popup / permission request / stop), generate AI character voice lines, and upload & manage your own sound library — all set with a few clicks, no code needed; floating desktop pet (default) or round ball, 3 appearances, dual-persisted config.

🐋 Character reference: a fan-made project for Angelina from Arknights. The bundled sample sounds are Angelina's "hirari do～", "Huh?" and "AWAWA!" voice clips, for personal learning and entertainment only — not for commercial use.

Install

Both methods work for either user type — Desktop users and npx dsh web users can use whichever they prefer.

Option A — Setup installer (recommended, zero dependencies, works for both):

If dsh isn't running, start it once (the Desktop app or npx dsh web) so its profiles get initialized — then quit it (the Desktop app or the dsh web server)
Download dsh-sound-lab-Setup-1.3.2-x64.exe and double-click it
The installer restarts dsh automatically — the floating Angelina desktop pet (default) means it's installed (falls back to the round ball automatically when no frames are bundled)

Update: re-download the latest Setup exe and run it (restarts dsh automatically; older versions are overwritten automatically — no need to uninstall first, no conflicts)
Uninstall: double-click dsh-sound-lab-UnSetup-1.3.2-x64.exe — it restarts dsh automatically when done

Note (just how it works — no action needed): the installer deploys the plugin to $DSH_HOME/plugins/dsh-sound-lab and registers it into every initialized profile (web, desktop, …) via the official dsh plugin command, so all profiles share the same copy. If no dsh CLI is found, it auto-detects one (desktop bundled runtime → system npx → downloads Node.js); if dsh hasn't been initialized yet (no profiles), it runs dsh web once to initialize. On finish it restarts dsh automatically: for Desktop it closes the running instance first, then launches it (exactly like double-clicking the app — no terminal window, unaffected by closing the installer); when no Desktop app is detected (web environment) it prompts you to restart the dsh web service manually. Fully automated, no manual config edits.

Option B — dsh CLI (needs Node.js, works for both):

If dsh isn't running yet, start it once first (npx @deepseek-ai/dsh web) so its profiles get initialized (npm run setup and the Option A installer do this automatically when uninitialized). Don't have dsh yet? Download the DeepSeek Harness Desktop app, or install Node.js (https://nodejs.org) and run npx @deepseek-ai/dsh web.

Pick the one command that matches how you run dsh — Desktop or web — and run only that one (choose 1 of 2):

# Desktop users —— run this one
npx @deepseek-ai/dsh plugin --profile desktop add dsh-sound-lab --config.minimumReleaseAge=0

# web users (dsh runs via `npx dsh web`) —— run this one
npx @deepseek-ai/dsh plugin --profile web add dsh-sound-lab --config.minimumReleaseAge=0

Then restart dsh (or just start it if it isn't running):

npx @deepseek-ai/dsh web

Development convenience: npm run setup auto-detects all local profiles and registers this plugin into each of them; npm run setup:deploy deploys a copy to $DSH_HOME/plugins and points every profile at it (release/fixed-use mode). Add --profile <name> (e.g. node tools/install.mjs --profile web --unify) to operate on a single profile only; node tools/install.mjs --npm --start installs from the npm package and auto-starts the matching side when done (Desktop window for desktop / web service + browser for web).

Features

Floating desktop pet (default, Arknights Angelina animation): idle (default standing) / greeting on click / move while dragging / docks at left·right·top·bottom edges (peek left·right / sleep top·bottom) / falls asleep after 90s idle / staff idle while a session runs / celebrates on session end / confused on options popup & permission request / falls down on stop & errors (holds the last frame, recovers on interaction); draggable, edge-docking, click to open the settings dialog; position is persisted, defaults to the left side
- Docking & hiding: top dock flips the pet upside-down and hides 80% (only 20% visible), bottom dock doesn't hide at all, left/right docks hide 40% and rotate 30° toward the screen center (clockwise left / counter-clockwise right); docking requires getting very close (30px) to the edge
- Animation & transitions: 60fps sprite-sheet playback (at half speed); event switches cross a 150ms half-fade (down to 40% opacity and back) with no overlap or ghosting
- Floating-window settings dialog: Theme (Angelina pet / built-in ball) + Event · Asset · Audio (each event can pick which sprite action and which sound to play — default: greeting →「呢？」, the rest silent; assets can be any loaded action, audio plays once per trigger while video can loop; when the matching trigger is enabled in "Trigger Conditions · Sounds", that detailed config wins); falls back to the round ball while frames load / on load failure
- Round ball (switchable): drag it anywhere; dragging it to a screen edge collapses it into a small half-ball (a ">" icon only); click to open the settings dialog
- Floating style (Appearance): desktop pet by default, switchable back to the round ball (auto-fallback to the ball when no frames are bundled)
Settings dialog: draggable (grab the title bar), z-index on top
- 4 trigger conditions: session end / options popup / permission request / stop, each with an independent 【enable checkbox + sound dropdown】offering built-in chime / no sound / a specific sound ("built-in chime" is a Web Audio arpeggio — no audio file required)
- Attention events (options popup / permission request) always ring — they play as soon as they appear, regardless of the conversation's running/viewing state, and take priority over the completion sounds
- Appearance: Whale Girl (default) / Pure White / Pure Black, plus a customizable voice name (the theme applies to every dialog at once and switches instantly)
- Volume slider (0–100%), test sound dropdown (incl. a "built-in chime" option) + ▶ preview + status bar, reset button position
- Sound library (local audio in the plugin sounds/ folder, managed via a sounds.json control file) + refresh + Upload dialog: pick/drag-drop upload, delete sounds, restore hidden bundled sounds; non-bundled sounds can be renamed via ✎
- "AI Voice" dialog: above the sound library — bring your own Alibaba Cloud Bailian API Key and cloned voice ID, type a line and generate speech in your custom voice (custom file name / preview / delete, one-click add to the library with automatic registration); a step-by-step tutorial with screenshots is included (click an image to zoom to 3× and drag to pan); token costs are on you
- 3 bundled sounds ("hirari do～" / "Huh?" / "AWAWA!") — cannot be deleted or renamed (delete = soft-hide, restorable anytime); hidden sounds are directly previewable with instant playback
- The sound library scrolls vertically beyond 4 entries and never exceeds the settings panel height; file extensions are hidden from sound names
Sound source: local audio files in the plugin sounds/ directory (mp3/wav/ogg/m4a/flac/opus/aac/wma/webm), served by the host side via the /dsh-sounds-control static server (Range/206 chunking, ETag, streaming)
Persistent config: dual-write to localStorage + host-side config.json, survives restarts; every field is sanitized (type/range/enum) on load, bad values fall back to defaults
Preload on startup: fetches config and sound list and buffers the configured sound when the app opens, so the first play is instant
Defaults: volume 60%; session end →「hirari do～」, options popup / permission request →「呢？」, stop → built-in chime
Falls back to a Web Audio built-in beep when no sound is selected or loading fails
Dev tooling: npm test (Node smoke tests for host-side list/config/upload/delete and browser-side logic), npm run setup / setup:fix / setup:deploy (multi-profile auto-config), releases/build-single-exe.ps1 (builds the Setup/UnSetup installers)

Directory structure

dsh-sound-lab/
├── package.json          # Package manifest (dsh.client / dsh.bundle.patch / types / scripts)
├── cordis.patch.yml      # Composition patch: mounts line ui-event-sounds
├── CHANGELOG.md          # Version history
├── README.md / README.zh.md
├── LICENSE
├── sounds/               # Sound library: bundled sounds (hirari do～/Huh?/AWAWA!) + uploaded/AI-generated audio
├── frames/               # Pet frame assets: <action>/sprite.webp sprite sheet (WebP lossless) + motion.json (60fps, 12 actions)
├── lib/
│   ├── index.js          # Host side: /dsh-sounds-control static service (list/config/audio/TTS) + /dsh-frames-control (pet frames)
│   ├── client.js         # Browser side: desktop pet / floating ball + settings dialog + trigger detection + playback
│   ├── tutorial/         # Illustrated tutorial images for AI voice generation (image1.png / image2.png)
│   └── types/index.d.ts  # Type declarations
└── tools/
    ├── install.mjs       # Multi-profile auto-config (setup / --fix / --unify / --deploy)
    ├── test-host.mjs     # Host-side API smoke tests (list/config/upload/delete/TTS)
    ├── test-client.mjs   # Browser-side logic smoke tests
    └── api/              # Standalone TTS script (tts_api.py + guide + reference audio)

Dual-side structure

The plugin is two halves, loaded automatically from a single install:

Host side (Node half) lib/index.js: runs in the DSH main process (Node), exposes the plugin sounds/ directory to the browser side through the /dsh-sounds-control static server (sound list / audio files / config.json persistence), and the frames/ pet assets through /dsh-frames-control (action list / frame PNGs)
Browser side (Web GUI half) lib/client.js: runs in the DSH Web GUI page — desktop pet (animation / drag / dock) or floating ball, settings dialog, trigger detection, and sound playback

Mounting of both halves is driven by package.json — no separate install needed:

dsh.client declaration (exports "./client") → loads the browser side in the Web GUI
dsh.bundle.patch (cordis.patch.yml) → registers the host side in the DSH main process

Because dsh plugins are isolated per profile ($DSH_HOME/profiles/<name> each has its own package.json / node_modules), a plugin must be registered in every profile you want it in — that's what the installer and tools/install.mjs automate (see Install).

Usage

Add sounds: during development, put audio files in the repo sounds/ folder (see the directory tree above) — they are reconciled into the library automatically on startup. When installed, there is no need to hunt for the folder — open the settings dialog → Sound library → Upload, then pick a local file or drag & drop to import (it is copied into the sounds/ folder of the install location automatically). You can also delete sounds from the list (bundled sounds are hidden, restorable anytime) or rename them via ✎.
Open the settings dialog: click the desktop pet (or the round ball)
Refresh the sound list: click "Refresh"; the plugin enumerates all audio files under sounds/
Configure triggers: for the four events (session end / options popup / permission request / stop), set 【enable + sound】independently — each dropdown offers "built-in chime / no sound / a specific sound" ("built-in chime" is a Web Audio arpeggio — no audio file required)
Test sound: pick one in the "Test sound" dropdown (which includes a "built-in chime" option) and click ▶ to preview; selecting "built-in chime" or nothing previews the Web Audio arpeggio. Adjust volume with the slider (0–100%)
In effect: the chosen sound plays automatically on matching events; a built-in beep plays as a fallback when nothing is selected or loading fails

Tip: drag the desktop pet anywhere; it docks at screen edges (peek left/right, sleep top/bottom — left/right hidden 40% with a 30° rotation, top flipped upside-down hidden 80%, bottom not hidden), falls asleep after 90s idle, greets on click; switch between desktop pet / round ball and Whale Girl / Pure White / Pure Black under "Appearance"; the floating-window settings dialog lets you adjust Theme / Event · Asset · Audio independently; position and config are saved automatically and survive restarts.

Disclaimer

This is a fan-made (unofficial) personal project with no affiliation, sponsorship, or authorization from the official Arknights team or Shanghai Hypergryph Network Technology Co., Ltd.
Character images, names, quotes, and voice assets referenced in this project (including Angelina's "hirari do～", "Huh?" and "AWAWA!" voice clips) belong to the official Arknights team and their respective right holders; the copyright of the voices belongs to the respective voice actors.
This project is for personal learning, research, and entertainment only — not for commercial use and not for profit.
The sound assets shipped with the project are local audio files added by users; users are responsible for ensuring their usage complies with applicable laws and the original right holders' requirements.
If any right holder believes any content of this project infringes their rights, please contact the author to remove the material and we will handle it promptly.
This project is provided as-is; the author is not responsible for any consequences arising from its use.