VoiceStudio：本地运行的 ElevenLabs 替代品，16 个 TTS 引擎，646 种语言

本地运行的语音工作台

ElevenLabs 的核心能力——声音克隆、视频配音、Audiobook 生成——你现在可以在自己的机器上跑，不需要账号、API Key、订阅，也没有用量计费。

VoiceStudio（曾用名 OmniVoice-Studio）是一个开源、全本地的语音工作台，把 16 个 TTS 引擎和 11 个 ASR 引擎统一到一个桌面应用里，支持 646 种语言，覆盖语音 AI 的六个核心工作流。

⭐ 12,712，fork 1,969，AGPL-3.0，Python + Bun/Tauri 构建，持续更新到 2026 年 8 月。

---

六个工作流

工作流	能做什么
声音克隆	3-15 秒参考片段，零样本克隆目标声音
声音设计	从年龄、口音、音调、风格、表达方式描述生成全新声音
视频配音	转录 → 翻译 → 保持说话人 → 合成 → 导出视频
转录 / 听写	系统全局快捷键，实时转录，可选本地 LLM 润色
故事与 Audiobook	多声音脚本、EPUB/PDF 导入、章节渲染、.m4b 导出
批量队列	大规模音频和视频任务并行处理，逐任务进度追踪

核心数据路径是本地的——音频和文字不经过任何第三方服务器。联网功能（远程 worker、模型下载）是明确的可选项，不是默认行为。

---

引擎生态：16 TTS + 11 ASR

VoiceStudio 的竞争优势不是一个单一模型，而是把当前最好的开源语音模型统一到一个界面里，按需安装、随时切换（Ctrl/Cmd+E）。

TTS 引擎（16 个）

引擎	语言数	声音克隆	macOS ARM	许可
VoiceStudio（默认，基于 k2-fsa/OmniVoice）	600+	✅	MPS	AGPL-3.0 / Apache-2.0
OmniVoice GGUF	600+	✅	MPS/CPU	AGPL-3.0 / Apache-2.0
CosyVoice 3	9+18方言	✅	CPU	Apache-2.0
GPT-SoVITS	5	✅	—	MIT
VoxCPM2	30	✅	MPS	Apache-2.0
IndexTTS 2.5	ZH/EN/JA/ES/AR	✅	CPU	Bilibili 模型许可
MLX-Audio	模型相关	部分	MLX	各异
MOSS-TTS-Nano	20	✅	CPU	Apache-2.0
Sherpa-ONNX	20+	—	CPU	Apache-2.0
KittenTTS	英语	—	CPU	MIT
PocketTTS	6种欧洲语言	✅	CPU	CC-BY-4.0（需授权）
Supertonic 3	31	—	CPU	OpenRAIL-M
MOSS-TTS-v1.5	31	✅	CPU	Apache-2.0
dots.tts	24	✅	CPU	Apache-2.0
Confucius4-TTS	14	✅	CPU	Apache-2.0
MOSS-TTS-v1.5	31	✅	CPU	Apache-2.0

默认引擎 VoiceStudio（OmniVoice）支持 600+ 语言、声音克隆、指令式合成，是开箱即用的最全能选项。没有克隆能力的引擎在视频配音和固定声音批量任务里会被直接拒绝（而不是悄悄换引擎），确保输出的可预期性。

ASR 引擎（11 个）

涵盖 Whisper 系列、WhisperX、Pyannote 说话人分离、实时流式识别等，配合 TTS 引擎构成完整的语音处理管线。

---

技术架构

层	技术
前端 / 桌面	Tauri + Bun（TypeScript）
后端	Python（uv 管理依赖）
计算	CUDA · Apple Silicon MPS/MLX · ROCm · CPU
接口	本地 REST/SSE/WebSocket API · OpenAI 兼容音频 API · MCP Server
模型管理	内置 Model Catalogue，在线安装/卸载/路由，支持远程 Worker

MCP Server 是一个值得关注的细节：VoiceStudio 暴露合成和转录工具给任何 MCP 客户端（Claude、Cursor 等），意味着你可以在 AI 编码工具里直接调用本地语音合成——不用离开工作区。

---

对比 ElevenLabs

	VoiceStudio	ElevenLabs
数据路径	本地（音频和文字不出机器）	经过 ElevenLabs 服务器
费用	免费（你提供算力）	订阅 / 按用量计费
离线使用	✅（模型下载后）	❌
引擎选择	16 个 TTS + 11 个 ASR	闭源，固定
语言支持	646 种（取决于引擎）	32 种
定制性	开源、可改引擎、可改路由	有限
维护	你自己管更新和算力	供应商管基础设施

适合 VoiceStudio 的场景：私有数据（法律/医疗/企业内容）、离线环境、高频批量生产（不想按量付费）、自研工作流集成。

---

硬件需求

	最低	推荐
OS	Windows 10 x64 · macOS 13.3+ Apple Silicon · Linux x86_64	当前系统版本
RAM	8 GB	16 GB+
磁盘	10 GB	20 GB+ SSD
GPU	可选（支持纯 CPU 模式）	NVIDIA CUDA 或 Apple Silicon
VRAM	4 GB（使用 GPU 时）	8 GB+（大型引擎更多）

注意：Intel Mac 无法运行本地 Python 后端，只能连接远程 Worker。Apple Silicon 是 macOS 上的原生平台。

---

安装与快速开始

从 GitHub Releases 下载对应平台的安装包（DMG / MSI / AppImage），首次启动自动创建 Python 环境并下载默认模型，后续启动复用缓存。

首次声音克隆三步：

1. 打开 VoiceStudio → Voice Cloning
2. 上传干净的参考音频（3 秒可用，5-15 秒效果更好）
3. 输入文字，选语言，点 Generate

从源码运行：

git clone https://github.com/debpalash/VoiceStudio.git
cd VoiceStudio
bun install
bun run desktop       # 桌面端
# bun run dev         # 浏览器 UI

---

总结

VoiceStudio 做了一件看起来简单但执行门槛很高的事：把语音 AI 的主流开源模型（16 个 TTS、11 个 ASR）统一到一个本地桌面应用里，覆盖从声音克隆到 Audiobook 生产的完整工作流，并暴露 OpenAI 兼容 API 和 MCP Server 给工具链集成。

12k Star，1.9k Fork，ElevenLabs 的替代品——不是功能对等，而是本地优先、数据自控、不按量计费。

GitHub: debpalash/VoiceStudio ⭐12712
官网: voicestudio.sh
Discord: discord.gg/bzQavDfVV9
许可: AGPL-3.0

VoiceStudio: Local-First ElevenLabs Alternative

ElevenLabs’ core capabilities — voice cloning, video dubbing, audiobook generation — can now run on your own machine, with no account, API key, subscription, or usage meter.

VoiceStudio (formerly OmniVoice-Studio) is an open-source, fully-local voice workstation that unifies 16 TTS engines and 11 ASR engines in a single desktop app, supports 646 languages, and covers six core voice AI workflows.

⭐12,712, 1,969 forks, AGPL-3.0, Python + Bun/Tauri, actively updated through August 2026.

---

Six Workflows

Workflow	What it does
Voice Cloning	Zero-shot clone from a 3–15 second reference clip
Voice Design	Generate a new voice from age, accent, pitch, style, and delivery instructions
Video Dubbing	Transcribe → translate → preserve speakers → synthesize → export video
Dictation	System-wide shortcut, live transcription, optional local LLM cleanup
Stories & Audiobooks	Multi-voice scripts, EPUB/PDF import, chapter rendering, .m4b export
Batch Queue	Large-scale audio and video job processing with per-job progress tracking

The core data path is local — audio and text never reach a third-party server. Network-backed features (remote workers, model downloads) are explicit opt-ins, not defaults.

---

Engine Ecosystem: 16 TTS + 11 ASR

VoiceStudio’s competitive advantage isn’t a single model — it’s unifying the best open-source voice models into one interface with on-demand installation and instant switching (Ctrl/Cmd+E).

TTS Engines (16)

The default engine — VoiceStudio (powered by k2-fsa/OmniVoice) — supports 600+ languages, voice cloning, and instruction-driven synthesis. Engines without cloning support are rejected rather than silently swapped in dubbing and pinned-voice batch jobs, keeping outputs predictable.

Key highlights:

• OmniVoice / OmniVoice GGUF — 600+ languages, clone + instruct, Apple Silicon MPS support
• CosyVoice 3 — 9 languages + 18 Chinese dialects, clone + instruct
• GPT-SoVITS — MIT, 5 languages, popular for high-quality clone
• IndexTTS 2.5 — Chinese/English/Japanese/Spanish/Arabic
• MLX-Audio — MLX native on Apple Silicon
• Sherpa-ONNX — lightweight, 20+ languages, CPU-first

ASR Engines (11)

Covers Whisper variants, WhisperX with speaker diarization (Pyannote), real-time streaming recognition, and more — a full audio processing pipeline alongside TTS.

---

Technical Architecture

Layer	Technology
Frontend / Desktop	Tauri + Bun (TypeScript)
Backend	Python (uv-managed)
Compute	CUDA · Apple Silicon MPS/MLX · ROCm · CPU
Interfaces	Local REST/SSE/WebSocket API · OpenAI-compatible audio API · MCP Server
Model management	Built-in Model Catalogue: install/remove/route, remote worker support

MCP Server is worth highlighting: VoiceStudio exposes synthesis and transcription tools to any MCP client (Claude, Cursor, etc.), letting you call local voice synthesis directly from within AI coding tools — without leaving your workspace.

---

Compared to ElevenLabs

	VoiceStudio	ElevenLabs
Data path	Local by default	Processed by ElevenLabs servers
Cost	Free (you supply compute)	Subscription or metered API
Offline	✅ after model download	❌
Engine choice	16 TTS + 11 ASR	Closed-source, fixed
Language support	646 (engine-dependent)	32
Customization	Open source, swap engines, modify routing	Provider-limited
Maintenance	You manage updates and compute	Provider manages infrastructure

Best fit for VoiceStudio: private data (legal, medical, enterprise), offline environments, high-volume batch production (no per-character billing), and custom toolchain integration.

---

Hardware Requirements

	Minimum	Recommended
OS	Windows 10 x64 · macOS 13.3+ Apple Silicon · Linux x86_64	Current supported OS release
RAM	8 GB	16 GB+
Disk	10 GB	20 GB+ SSD
GPU	Optional (CPU mode supported)	NVIDIA CUDA or Apple Silicon
VRAM	4 GB if using GPU	8 GB+ (large engines need more)

Note: Intel Macs cannot run the local Python backend — connect a remote worker instead. Apple Silicon is the native macOS platform.

---

Install and Quick Start

Download the installer for your platform from GitHub Releases (DMG / MSI / AppImage). First launch creates a managed Python environment and downloads the default model; subsequent launches reuse the cache.

First voice clone in three steps:

1. Open VoiceStudio → Voice Cloning
2. Add a clean reference clip (3 seconds works; 5–15 seconds usually better)
3. Enter text, choose a language, click Generate

Run from source:

git clone https://github.com/debpalash/VoiceStudio.git
cd VoiceStudio
bun install
bun run desktop    # desktop app
# bun run dev      # browser UI

---

Summary

VoiceStudio does something that sounds simple but has a high execution bar: it unifies the major open-source voice models (16 TTS, 11 ASR) into a local desktop app covering the complete workflow from voice cloning to audiobook production, then exposes an OpenAI-compatible API and MCP Server for toolchain integration.

12k stars, 1.9k forks. An ElevenLabs alternative — not feature-parity in every edge case, but local-first, data-controlled, and no per-character billing.

GitHub: debpalash/VoiceStudio ⭐12712
Website: voicestudio.sh
Discord: discord.gg/bzQavDfVV9
License: AGPL-3.0