Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Voice, Speech & Multimodal

YoVoice Review: Local Voice Cloning and Music Studio

·7 min readBy · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

YoVoice (version 0.8.0) is an Apache-2.0 desktop studio that combines text-to-speech, voice cloning, AI song generation and score writing, and its maker says generation runs offline on your own computer. It also ships a CLI, a local HTTP API and an MCP server for agents, and every statement here comes from the maker's website, documentation and GitHub repository.

Key Takeaways

  • Maintenance: the main branch last changed on 17 November 2026 and version 0.8.0 was published on 15 November 2026, so no stale-project warning applies.
  • Cost: free under the Apache-2.0 license, with no paid plan found; many downloadable model weights carry their own terms.
  • Models: the README says all voice, music and transcription models are downloaded inside the app or with the CLI.
  • Platforms: installers for macOS, Windows and Linux, plus a standalone CLI archive for each.
  • Maker: an individual developer, leemysw; no company imprint was found.

📍 In One Sentence

YoVoice is an open-source desktop studio that turns text into speech, clones and designs voices, generates songs and scores with local models and gives AI agents a CLI, an API and an MCP server.

💬 In Plain Terms

Think of it as a voice booth and a small music studio in one program, with the work done on your own machine instead of a subscription service.

📌Note: This review is based on yovoice.leemysw.com, the GitHub repository, its documentation and the release page, checked on 17 November 2026. PromptQuorum has not installed or run the app, so voice quality, music quality and speed are not assessed.

YoVoice (written "yovoice" by its maker) is an open-source desktop studio that generates speech, clones and designs voices, and writes songs and scores with models running on your own computer. It also offers a command line, an HTTP API and an MCP server for AI agents, and the app is free under the Apache-2.0 license. This review is based on the website, the GitHub repository, its documentation and the release page, checked on 17 November 2026; PromptQuorum has not tested the app hands-on.

YoVoice Review: Local Voice Cloning and Music Studio

What Is YoVoice?

YoVoice is a local-first audio studio: it turns text into speech, designs or clones a voice, generates songs and instrumentals and arranges multi-track scores, using models you download and run on your own computer. The README calls it an open-source studio for macOS, Windows and Linux; the project spells its name "yovoice", and its website is Chinese first with an English version.

The inference engine is audio.cpp, a native C++ engine by ShugoAI, so the README says no Python or Conda setup is needed. Version 0.8.0 added community song sharing, customizable keyboard shortcuts, six themes and new speech-recognition models, according to the changelog.

Get It

YoVoice is a download for macOS (Apple silicon), Windows and Linux that installs like any other app; a standalone CLI archive is also published, and no terminal is needed for the desktop app.

Option
How to get it
macOS.dmg for Apple silicon, macOS 14 or later: GitHub Releases
WindowsSetup .exe for x64, with CUDA 12.4 and 13.3 variants: GitHub Releases
Linux.deb (Debian/Ubuntu) or .tar.gz for x64: GitHub Releases
CLIStandalone zip per OS: CLI guide
Websiteyovoice.leemysw.com
Privacy policyNone found; only community terms exist
Source codegithub.com/leemysw/yovoice, Apache-2.0 license

Version reviewed: 0.8.0, published on 15 November 2026 per the GitHub release page; the newest commit on the main branch dates from 17 November 2026. GitHub is only the delivery channel for these installers, and models are downloaded separately inside the app.

Requirements

No minimum RAM, VRAM or disk size is stated in the sources; the demanding part is the models, which are downloaded separately and differ widely in size.

Operating system

What the sources say:
macOS 14 or later on Apple silicon, Windows 10/11 x64, Linux x64; the Windows installer fetches the WebView2 runtime if needed and Linux needs WebKitGTK 4.1 (the .deb installs it)

Acceleration

What the sources say:
Metal on macOS; the Windows CUDA 12.4 and 13.3 installers bundle a runtime for NVIDIA GPUs; Linux runs on the CPU out of the box, with a Vulkan GPU runtime downloadable in Settings

Memory and VRAM

What the sources say:
Depends on the model; the docs list file sizes per model, for example IndexTTS 2.5 at 3.50 GB in Q8, and recommend a GPU for Breeze TTS 2

Models

What the sources say:
Downloaded inside the app or with the CLI, with pause and resume; the sources give no total download size

Data folder

What the sources say:
~/.yovoice by default, movable in Settings; uninstalling the app preserves it

Internet

What the sources say:
Needed to download models and runtimes; the docs say generation works offline once the engine and a model are installed

Backends are as listed in the README and CLI guide; PromptQuorum did not check which models run on which backend.

Features Confirmed by the Sources

Every item below comes from the README, the documentation or the changelog; none has been independently tested.

  • Speech, voice design and cloning. Clone a voice from a reference recording, design one from a text description and adjust emotion and delivery; the model list includes IndexTTS 2.0 and 2.5, VoxCPM2, Qwen3-TTS, XTTS v2 and Kokoro.
  • Story dubbing. Import text or subtitle files, assign a character to each line and generate line by line, with several takes per line to compare.
  • Music and scores. Songs with lyrics and instrumentals through ACE-Step 1.5 or YuE2, instrumentals and sound effects through Stable Audio 3, six-stem song separation, and General MIDI score arranging with MIDI and stem export.
  • Transcription. Audio or video becomes timed subtitle lines exported as SRT or WebVTT, using models such as Qwen3 ASR and Parakeet.
  • Agents and API. An installable Agent Skill and CLI, an authenticated HTTP API with an MCP endpoint through yovoice serve, a Local API inside the desktop app on 127.0.0.1, and an optional assistant panel that works with services such as DeepSeek, Qwen, OpenAI, Anthropic, Ollama and LM Studio.
  • Interface. Twelve interface languages, six themes and editable keyboard shortcuts.

Cost

The app costs nothing to download, and neither the website nor the repository lists a paid plan, trial or subscription.

App

What the sources say:
Free; only community sharing is described as needing an account

AI assistant

What the sources say:
Optional; a pay-as-you-go API or plan from the provider you connect, or a local Ollama or LM Studio service, which usually needs no key

Models

What the sources say:
The docs describe downloading models inside the app and mention no fee; licenses differ per model

No price was listed in any currency; the structured data on the website lists a price of 0 USD for the app.

Data Flow and Privacy

The README says generation runs entirely on your own CPU or GPU and that manuscripts, reference voices and creations never leave the computer, and the website repeats "100% offline privacy". The architecture document adds that apart from model and runtime downloads and community sharing a user starts, the only network use is the AI service you configure; the README separately describes background update checks on macOS and Windows.

The optional assistant sends your typed request and a fixed instruction straight to the provider you choose, and keys are stored in the local database with owner-only permissions. Community sharing, added in 0.8.0, uploads a preview audio file and a project package after you sign in with a leemysw account, and the community terms say sign-in records the terms version, the time and your network address.

📌Note: These are the maker's declarations, not audit results. PromptQuorum has not read the code or inspected the app's network traffic, and with no privacy policy to read, these statements come from the README and documentation only.

Voice Cloning, Consent and Model Licenses

Cloning needs only a reference recording, and the README and documentation pages we read describe no check that the voice belongs to the user, so consent is your responsibility. The maker's community terms forbid sharing a real person's voice without consent, including clones of public figures, actors and singers, and impersonation; those terms govern the community, not private use of the app.

  • Consent and law. Whether cloning a voice is lawful depends on the person, the purpose and the country; this review is not legal advice, and PromptQuorum has not assessed any jurisdiction.
  • Model licenses differ. The README says OmniVoice, YuE2, Breeze TTS 2, Voxtral 4B TTS and MuScriptor weights are restricted to non-commercial use, XTTS v2 and Drama 3 use their own Coqui and Fish Audio licenses, IndexTTS models have a separate model license and Stable Audio 3 is free below USD 1M annual revenue under the Stability AI Community License. The Apache-2.0 license covers the app, not those weights.
  • Sharing rules. The community terms require a sharer to state a voice's source truthfully (own voice, another person with clear consent, or a synthetic voice) and to choose a non-commercial license for works from models that forbid commercial use.

What We Could Not Verify

  • Output quality and speed. PromptQuorum did not listen to speech or music, measure generation speed or test cloning accuracy on any model; the website offers audio samples, which we did not evaluate.
  • Project maturity. The repository was created on 15 September 2026 and is published under one developer's name, so long-term maintenance is unknown.
  • Take retention. The Unreleased section of the changelog says a fix stops generation from silently deleting older takes after ten versions; we did not check whether version 0.8.0 behaves that way, so keep copies of takes you need.
  • Language coverage. Supported speech languages depend on the model and the sources give no single figure; we tested no language.
  • Installer signing. Whether the macOS and Windows installers are notarized or code-signed was not checked.

Competitors and Alternatives

Voicebox

Platforms:
macOS, Windows (Linux from source)
Price / license:
Free / MIT
Key difference:
Voice cloning, text-to-speech and Whisper dictation with an MCP server; its review lists no music generation or score writing
Articles about Voicebox (1)

Also mentioned in:

XTTS v2

Platforms:
macOS, Windows, Linux
Price / license:
Free / CPML
Key difference:
A single cloning model used as a library, with no studio, music or agent features
Articles about XTTS v2 (9)

+6 more not shown

These are same-category apps from the PromptQuorum directory; confirm each app's current price, license and platforms on its own page.

Compare in directory

Frequently Asked Questions

Can I run it from a script or an AI agent?

Yes: the CLI works without the desktop app, and an installable Agent Skill lets an agent set up the CLI, engine and model. The server and the in-app Local API use a preconfigured bearer token rather than an OAuth login, and the Local API is reachable only from the same computer.

Which model should I start with?

The Quick Start suggests Kokoro-82M for fast narration or VoxCPM2 for cloning, and the docs say Kokoro has built-in voices and no cloning. We did not compare output quality.

Does it support my language?

The interface has twelve languages, but speech depends on the model: the docs say IndexTTS 2.5 adds Japanese, Spanish and Arabic to Chinese and English, XTTS v2 covers 17 languages and Audio8 TTS Realistic 11. We tested none of them.

Can the AI assistant run fully locally?

It can connect to Ollama or LM Studio on the same computer, per the docs, which add that larger models write noticeably better arrangements and small local models may fail the format check.

Do shared songs include my reference voices?

No: the community terms and architecture document say a shared package holds that version's audio, lyrics and generation settings but not reference voices, a person reviews each work before it is public, and shared works can be deleted in the app's settings.

Verdict

YoVoice is a young but actively developed, Apache-2.0 way to generate speech, clone voices and make AI music on your own computer, with a CLI and an MCP server that let agents drive it. Its offline statements come from the README and documentation rather than an independent audit, the repository is only two months old with a single named developer, several model weights forbid commercial use, and the Mac build covers Apple silicon only. It suits creators and developers who want one free local audio tool and will download large model files; anyone who needs proven output quality, commercial-use clarity for each model or a vetted data policy should test it and read each model's license first.

Sources

← Back to Power Local LLM

MCP