dsh-pet-tts

English | 中文

Speak the dsh-pet pet's speech bubbles aloud using Microsoft Edge's online neural voices. The browser half watches the .dsh-pet-bubble DOM node; the host half synthesizes mp3 through a private Python virtualenv running the open-source edge-tts client.

Not a general-purpose "read the assistant's reply" plugin — it is scoped to the pet's own bubbles, so it stays quiet unless the pet actually says something. If you want the assistant's messages read instead, use one of the reply-reading plugins in the market.

Requirements

• DeepSeek Harness with a web UI (dsh web / the desktop app's web view). Tested on 0.2.1-rc.5.
• dsh-pet for the bubbles (optional — the console API below works even without it).
• Python 3.9+ on PATH, or point DSH_PET_TTS_SYSTEM_PYTHON at an interpreter.
• Network access to speech.platform.bing.com, and once to PyPI (or a mirror) for the one-time edge-tts install.
• Node.js 20+.

Install

dsh plugin --profile <your-profile> add github:jimmyfenwick/dsh-pet-tts

or from a release tarball:

dsh plugin --profile <your-profile> add https://github.com/jimmyfenwick/dsh-pet-tts/releases/latest/download/dsh-pet-tts.tgz

Restart or reload the host, then click once inside the page (browsers block audio before a user gesture) and let the pet say something.

On first synthesis the plugin creates a virtualenv at ~/.dsh/dsh-pet-tts/venv and installs edge-tts into it — roughly 30 MB, one time, nothing global. Progress appears in the host log as [dsh-pet-tts] creating virtualenv … / installing edge-tts ….

Presets

id	voice	rate	pitch	volume	feel
anime-girl (default)	zh-CN-XiaoyiNeural	+35%	+38Hz	+10%	bright, fast, anime-girl
anime-girl-hi	zh-CN-XiaoyiNeural	+35%	+50Hz	+10%	higher, childlike
anime-girl-soft	zh-CN-XiaoyiNeural	+20%	+38Hz	+10%	same pitch, calmer pace
anime-boy	zh-CN-YunxiNeural	+20%	+15Hz	+5%	young male
netstar	zh-CN-XiaoxiaoNeural	+25%	+25Hz	+10%	short-video narrator
xiaoyi-raw	zh-CN-XiaoyiNeural	+0%	+0Hz	+0%	natural
xiaoxiao-raw	zh-CN-XiaoxiaoNeural	+0%	+0Hz	+0%	natural, general purpose

The pitch/rate values mirror the ones shipped by dsh-pet-dfeiyu-mo, so the same voices sound familiar.

Console API

dshPetTTS.setPreset('anime-girl-hi')          // switch preset, persisted in localStorage
dshPetTTS.say('主人，我回来了')                 // speak one line immediately
dshPetTTS.say('你好', { voice: 'en-US-AriaNeural', rate: '+10%' })
dshPetTTS.mute(true)                          // silence; mute(false) to resume
await dshPetTTS.voices()                      // all Microsoft voices known to the host
await dshPetTTS.health()                      // { ready, python, edgeTts, dataRoot }

Routes

The host exposes a prefix route, useful for scripting or a custom frontend:

GET /dsh-pet-tts/ping                          → { ok, defaultPreset, presets }
GET /dsh-pet-tts/health                        → { ready, python, edgeTts, dataRoot, venv }
GET /dsh-pet-tts/presets                       → full preset table
GET /dsh-pet-tts/voices                        → every voice Microsoft currently returns
GET /dsh-pet-tts/say?text=…&preset=…           → audio/mpeg
   extras: voice=, rate=, pitch=, volume=

Synthesized audio is cached under ~/.dsh/dsh-pet-tts/cache keyed by text + voice + parameters, so repeated lines cost nothing.

Configuration

env var	meaning
DSH_PET_TTS_ROOT	data directory (default $DSH_HOME/dsh-pet-tts)
DSH_PET_TTS_PYTHON	use this interpreter as-is, skip the virtualenv entirely
DSH_PET_TTS_SYSTEM_PYTHON	base interpreter used to create the virtualenv

If pip install cannot reach PyPI, the plugin retries the Tsinghua mirror first, then upstream PyPI; set DSH_PET_TTS_PYTHON to a pre-provisioned interpreter to skip installation altogether.

Limitations

• Voice text is capped at 300 characters per bubble; longer lines are truncated, because a pet bubble is not a paragraph.
• Only DOM nodes with the dsh-pet-bubble class are spoken.
• The first sound waits for a click/keypress — browser autoplay policy, not a plugin bug.
• Synthesis needs the network; there is no offline model.
• Built for macOS/Linux. Windows code paths exist (Scripts/python.exe) but are untested.
• Uses Microsoft's public Edge read-aloud endpoint. It is undocumented, unaffiliated with this project, and can change or throttle without notice; check Microsoft's terms before commercial use.

Uninstall

dsh plugin --profile <your-profile> remove dsh-pet-tts

Optionally delete ~/.dsh/dsh-pet-tts (virtualenv + cache, ~30 MB).

License

MIT.

---

中文

把 dsh-pet 桌宠的气泡台词念出来，声音走微软 Edge 的在线神经音色。前端监听 .dsh-pet-bubble 节点，宿主侧用私有 Python 虚拟环境里的开源客户端 edge-tts 合成 mp3。

它只念桌宠气泡，不念助手正文——所以桌宠不说话时它是安静的。想把助手回复念出来，请装市场里那类"朗读回复"插件。

依赖

• 带 Web UI 的 DeepSeek Harness（dsh web 或桌面版的网页视图），在 0.2.1-rc.5 上验证。
• dsh-pet（用于产生气泡；不装也能用控制台 API）。
• Python 3.9+，或者用 DSH_PET_TTS_SYSTEM_PYTHON 指定解释器。
• 能访问 speech.platform.bing.com，以及一次性访问 PyPI（或镜像）装 edge-tts。
• Node.js 20+。

安装

dsh plugin --profile <你的档名> add github:jimmyfenwick/dsh-pet-tts

或用 release 里的 tarball：

dsh plugin --profile <你的档名> add https://github.com/jimmyfenwick/dsh-pet-tts/releases/latest/download/dsh-pet-tts.tgz

装完重启或刷新，先在页面里点一下（浏览器禁止用户手势前自动播音频），再让桌宠说话。首次合成时会在 ~/.dsh/dsh-pet-tts/venv 建虚拟环境并装 edge-tts——约 30MB，一次性，不装全局。宿主机日志里会看到 [dsh-pet-tts] creating virtualenv … / installing edge-tts …。

预设

见上表（anime-girl 默认）。参数与 dsh-pet-dfeiyu-mo 同源。

dshPetTTS.setPreset('anime-girl-hi')
dshPetTTS.say('主人，我回来了')
dshPetTTS.mute(true)
await dshPetTTS.voices()
await dshPetTTS.health()

配置与限制

同英文节：DSH_PET_TTS_ROOT、DSH_PET_TTS_PYTHON、DSH_PET_TTS_SYSTEM_PYTHON；每条气泡最多念 300 字；首次发声要等一次点击；合成需要联网；macOS/Linux 验证过，Windows 路径支持未实测；用的是微软公开朗读接口，未与之关联，接口可能随时变动，商业使用前请核对微软条款。

卸载

dsh plugin --profile <你的档名> remove dsh-pet-tts

如需要，再删 ~/.dsh/dsh-pet-tts（虚拟环境 + 缓存）。MIT 许可。