onkey README

项目的 README,来自 github.com/dsmurrell/onkey。合并为一个 markdown 文件:README.md。

OnKey

Learn to sing the songs you love. · singonkey.app

Saving shower singers from themselves since 2026.

OnKey's singing lane: note bars scroll towards the playhead and the singer's pitch is drawn over them, green where it's on the note
Me singing along to a song OnKey learned from one play. The full demo, with sound.

OnKey is a free, open-source singing trainer for the Mac. Play a song in Spotify, Apple Music, or YouTube in a browser, and sing along. The first time, you get the lyrics, and OnKey learns the singer's melody when the song ends. After that, the notes scroll towards you, your pitch is drawn over them, and you get a score with a report on every line.

OnKey doesn't provide, stream or download music. Your own music app plays it.

Install

You need a Mac with Apple Silicon (M1 or later) and macOS 14.2 or later.

Download OnKey.dmg (from the latest release), open it and drag OnKey to Applications.

Setup asks for the microphone, to hear you, and for system audio recording, to hear the music app. It also asks whether OnKey may follow Spotify and Apple Music: choose Listen only and it follows songs by their sound alone. Getting started walks through it, and the FAQ answers the usual questions.

How it works

When a song ends, OnKey learns it on your Mac: a model separates the lead vocal from the music, CREPE tracks the singer's pitch, and the GAME note model turns it into one bar per syllable. On later plays it finds its place from the music itself and tracks your voice live. Singing an octave down counts, and being a touch late isn't punished. How it works and Scoring have the details.

Privacy

OnKey never saves the music it listens to. It analyses the sound in memory as it plays. Your recordings and the songs it learns stay on your Mac (a take sung over speakers can contain some of the music). There's no account, no analytics and no crash reporting.

These requests do go out:

Settings has a switch for Shazam, for lyrics and for the update check. Privacy lists everything OnKey keeps and sends, and scripts/release-hosts.txt lists every web address in a release.

Using OnKey fairly

OnKey is for practising at home, with music you're entitled to play. Please:

Rights holders can reach me at [email protected] (LEGAL.md).

Help

Use Report a problem in the app (Settings, the Help menu, or a take's page). It saves a zip of numbers and logs on your Mac, with no audio and no recordings of you, for you to attach to an issue if you want to. Troubleshooting covers the common problems. Questions, or a song that's out of step? Ask in OnKey's Discord, or email [email protected].

Building from source

If you'd rather not trust my build, you can build your own. You need an Apple Silicon Mac with the Xcode command line tools, Rust, Node, and Python 3:

git clone https://github.com/dsmurrell/onkey && cd onkey
npm install && npm --prefix ui install
python3 -m venv .venv && .venv/bin/pip install -r scripts/requirements-models.txt
PYTHON=.venv/bin/python scripts/fetch-models.sh   # downloads the models and checks them by SHA-256
npx tauri build

Your build runs on your own Mac. The one thing it can't do is ask Shazam to name a new song, which needs my Apple signature. Spotify and Apple Music say what's playing themselves, so that only affects YouTube and Listen only. CONTRIBUTING.md has the details, the tests, and how to check a release's checksums. My releases are signed with my Apple Developer ID and notarised by Apple (spctl -a -vv /Applications/OnKey.app shows who signed it).

Licence and credits

MIT. See LICENSE. Modified versions can't be called OnKey or use its icon. If you build on OnKey's code, a credit linking back here is appreciated.

OnKey learns songs with CREPE, GAME, wav2vec 2.0 and UVR's MDX-Net Karaoke 2. The credits list each one, and THIRD_PARTY_NOTICES.md has the full notices and licence texts.

OnKey is made by Daniel Murrell.

onkey (dsmurrell/onkey)

正在搜索数据库…

428 ·GitHub
这是什么软件?OnKey 是一款免费、开源的 Mac 唱歌训练工具。你在 Spotify、Apple Music 或 YouTube 中播放一首歌,跟着唱。播放一遍之后,它就学会了歌手的旋律;之后再唱时,音符会朝你滚动过来,同时把你的音高叠画在上面,并给出分数。
这是原生 Mac 应用吗?它自己会播放音乐吗?这是一款原生 macOS 桌面应用,使用 Tauri 和 Rust 构建。它本身不提供、不串流,也不下载任何音乐。歌曲由你自己的 Spotify、Apple Music 或浏览器播放,OnKey 只是在旁边聆听。
我需要什么样的 Mac 才能运行它?需要搭载 Apple 芯片(M1 或更新机型)、运行 macOS 14.2 或更高版本的 Mac。不支持 Intel 芯片的 Mac,也不支持 Windows。
我能用它下载音乐,或者直接获取音乐本身吗?不能。OnKey 从不保存或提供它所分析的音乐,它只是聆听你自己的应用中正在播放的内容。如果是在唱歌活动中使用,请改用有授权的服务,而不要依赖 OnKey。
使用 OnKey 时,它会向外发送哪些数据?你的声音和音乐音频都不会离开你的 Mac。它只会发送一段简短的 Shazam 音频指纹,用来识别无法辨认的歌曲;会把歌名和歌手发送给 LRCLIB 获取歌词,发送给 iTunes 获取歌曲时长和封面;并且每天向 GitHub 检查一次更新。没有账户,也没有数据分析。
OnKey 究竟是怎么学会一首歌的旋律的?歌曲播放结束后,它会把主唱人声从伴奏中分离出来,用 CREPE 追踪歌手的音高,再用 GAME 音符模型把结果转换成每个音节对应的一条音符条。所有这些都在你的 Mac 上完成。
唱错八度,或者稍微晚一点,会被扣分吗?不会。比原曲低一个八度来唱,同样算数;稍微晚一点也不会被扣分,这是根据评分说明得出的。
我能分享自己的演唱录像吗?可以。使用内置的分享功能,它会生成一段短视频,展示你的声音和音高走势,不包含音乐音频。你可以把它发到 Instagram,并在 Instagram 自己的音乐库中添加一段有授权的歌曲音轨。
哪些 AI 模型负责旋律和人声的识别?OnKey 使用 CREPE 和 GAME 进行音高与音符检测,使用 wav2vec 2.0 以及 UVR 的 MDX-Net Karaoke 2 模型进行人声分离。
作者
dsmurrell
大小
7.8 MB
星标
428
时长
1个月
提交 · 90 天