midasheng-gen-mcp
midasheng-gen-mcp
从文本生成连贯的16 kHz 混合音频场景——语音、音乐、音效和环境音一次生成——由MiDashengLM-Gen(小米研究院)驱动,这是一个基于LLM的自回归流匹配模型(Qwen3-1.7B骨干网络,总计2.9B参数)。采用Apache-2.0许可,完全在您的GPU上运行。
本工具封装了什么
MiDashengLM-Gen——据论文所述,首个端到端训练的通用文本转音频模型:LLM驱动逐token流匹配,生成可变长度的音频场景,其语音清晰度接近专用TTS(Seed-TTS WER 12.15% -> 2.79%,而专用TTS为1.24%)。支持9种语言和情感控制。检查点从Hugging Face自动下载(约6 GB)。模型权重从不捆绑——请参阅docs/WRAPPEE.md。
Related MCP server: MMAudio MCP
您可以做什么
运行方式:一个本地FastMCP 3.4服务器(stdio用于Claude Desktop,HTTP /mcp用于Cursor/webapp),附带React仪表板。模型在首次生成时延迟加载,并保留在GPU上直到卸载。
方向 | 产物 | 备注 |
输入 | 结构化描述(描述/asr/语音/音效/音乐/环境视图),引导参数 + 种子参数 | 通过MCP工具、REST或webapp |
输出 | 16 kHz单声道WAV场景,索引于SQLite中,可浏览/导出 |
|
一次调用生成完整音景——人群笑声 + 爵士乐片段 + 喜剧演讲,或雨声 + 雷声 + 森林环境音
语音清晰度接近专用TTS质量,9种语言,通过
speech视图控制情感场景库支持分页、音频播放和webapp导出
异步任务API用于长时间生成;聊天中的预制UI卡片
通过学习到的停止头实现可变长度输出——无需固定时长裁剪
快速安装
最快的方式是使用Claude Desktop的.mcpb包(所有选项请参阅INSTALL.md):
从Releases下载
midasheng-gen-mcp-v0.1.0.mcpb将其拖放到Claude Desktop上
首次使用将自动下载约6 GB的检查点
或者克隆并双击start.bat以启动完整堆栈(后端 + webapp)。
示例提示
"生成一个喜剧俱乐部场景:一句妙语、人群笑声和爵士乐队片段" ->
audio_scene(operation="generate", caption="A comedian delivering a punchline followed by uproarious crowd laughter", asr="And that is why I never buy cheap luggage anymore!", speech="expressive comedic male voice", music="sudden upbeat jazz band sting", sfx="crowd laughter", env="intimate comedy club")"在森林中制作一个夜晚的雷暴" ->
audio_scene(operation="generate", caption="A rolling thunderstorm in a forest at night", sfx="distant thunder and heavy rain", env="dense forest", seed=42)"模型状态如何?" ->
audio_scene(operation="status")
文档
文档 | 内容 |
所有安装方法、前提条件 | |
首次运行模型下载、GPU检查、常见陷阱 | |
MiDashengLM-Gen论文、权重、许可、演示 | |
系统架构、端口、数据流 | |
环境变量、配置选项 | |
所有可用工具 | |
贡献指南、本地设置 | |
常见问题 |
系统要求
配备CUDA GPU的Windows/Linux/macOS(RTX 4090级别:约12 GB显存(fp32);CPU推理可行但速度慢)
通过uv安装的Python 3.12+,Node.js 20+,bun(在裸机上由
start.ps1自动安装)约10 GB可用磁盘空间(检查点约6 GB + torch堆栈)
许可
Apache-2.0(模型和本封装)。请参阅上游仓库中的使用限制部分:禁止非法/军事用途,禁止对未成年人或群体造成伤害。
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- FlicenseAqualityDmaintenanceEnables AI assistants to generate and control real-time audio synthesis through natural language descriptions using SuperCollider. Features 10 built-in synth types, pattern sequencing, audio recording, and server lifecycle management for creating sounds from simple English descriptions.111

MMAudio MCPofficial
AlicenseBqualityCmaintenanceEnables AI-powered video-to-audio and text-to-audio generation using MMAudio's API. Create synchronized audio from video content or generate audio from text descriptions with configurable parameters.3103MIT- Alicense-qualityCmaintenanceEnables users to generate sound effects from text descriptions using Meta's AudioGen model. Specifically designed for Apple Silicon Macs, it supports single and batch audio generation directly from natural language prompts.1MIT
- Alicense-qualityBmaintenanceGenerate and refine AI images/audio/video through natural conversation.397Apache 2.0
Related MCP Connectors
Generate images, video, music and voice from your CLI or AI agent. On-brand AI media toolkit.
AI-manageable audio CDN: upload, transcode, normalize, stream & deliver audio, plus grounded docs.
Transform video, audio and images, and generate media from prompts. FFmpeg, captions, models.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/sandraschi/midasheng-gen-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server