npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

dsh-plugin-wwrs-voice

v0.1.1

Published

Voice generation DSH plugin (open source): MiniMax TTS, VoxCPM2 and voice design behind one facade. Sole runtime dependency is @deepseek-ai/dsh-tools (defineTool DSL).

Readme

dsh-plugin-wwrs-voice

English

语音生成 DSH 插件(开源):MiniMax TTS、VoxCPM2 与音色设计收敛到同一立面之后。唯一运行时依赖是 @deepseek-ai/dsh-tools(defineTool DSL)。W1 已落地:同一立面后四个工具(voice_synthesize 全实现 MiniMax T2A;voice_clone / voice_design 标 experimental)。

工具

四个工具的读写都在已配置的工作区之下,并受写保护守卫覆盖。

| 工具 | 参数 | 语义 | | -------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------- | | voice_synthesize | text(超长拒收,请分段调)、voiceId、model、outputPath(工作区内 .mp3)、speed?(0.5–2)、emotion?、overwrite?(缺省 false) | MiniMax T2A 全实现 → mp3 并附词时间戳;voiceId/model 显式传入、不代选 | | voice_clone | text、refAudio(工作区内已存在音频)、outputPath、seed?、overwrite?(缺省 false) | experimental——v1 只做参数校验不合成,校验后指引调用方调对应 comfyui 模板 | | voice_design | text、instruct、outputPath、seed?、overwrite?(缺省 false) | experimental——v1 只做参数校验不合成,校验后指引调用方调对应 comfyui 模板 | | pronunciation_dict | — | 内置纠音与 <工作区>/.wwrs/pronunciation-dict.json(若有)的合并 |

合同

远端 MiniMax T2A 收敛到 voice_synthesize 门面之后(直调、落盘、词时间戳)。voice_clone / voice_design 不直调 Comfy——只做参数校验并指引调用方调 dsh-plugin-wwrs-comfyui 对应模板。凭据只走环境变量(MINIMAX_API_KEY;缺失即大声失败,无回落)。

配置

| 键 | 含义 | 缺省 | | ----------- | ------------------ | --------------------------------------- | | workspace | 项目根(绝对路径) | WWRS_WORKSPACE 环境变量,再中性锚探测 |

工作区解析:config.workspace > 环境变量 WWRS_WORKSPACE > 向上探测 .wwrs/workspace.json > 大声失败。请把 workspace 指向绝对路径的项目根,以便语音产物落到已知位置。初始 RPM 配额沿用老分档且可配置:合成 10/分、复刻 60/分、设计 20/分。重试缺省 2 次,只对 429/5xx 生效(4xx 与提供商业务码永不重试)。缺失即大声失败并给出路;绝不静默降级。

安装

{
  "dependencies": {
    "dsh-plugin-wwrs-voice": "link:/path/to/plugin-wwrs-voice"
  },
  "dsh": { "profile": { "bundles": ["dsh-plugin-wwrs-voice"] } }
}

将本包 link 安装进宿主 profile 并启用 dsh-plugin-wwrs-voice 包。重启宿主使 wwrs-voice 层挂载;空壳会在启动日志中打印工作区来源。

验证

跑 pnpm check(prettier 检查 + tsc --noEmit + node --test tests/*.test.ts)。

无浏览器半

纯服务端工具——pnpm check 即全部门,无 check:browser 链。

已知边界

  • voice_clone / voice_design 标 experimental:v1 只做参数校验并指引调用方调对应 comfyui 模板——本包不打出任何 Comfy 调用。
  • voice_synthesize 永不代选 voiceId/model;超长文本拒收(请分段调)。
  • 重试只对 429/5xx 生效(4xx 与提供商业务码永不重试)。
  • pronunciation_dict 只在工作区自定义文件存在时才合并。

许可

MIT.