Transcribe
Slow down, mark up, and clip audio for transcribing music by ear.
macOS
Apple Silicon and Intel · macOS 12 or later
Paste one line in Terminal, or download the app.
Open Terminal (Spotlight → “Terminal”), paste this line and press Return:
curl -fsSL https://transcribe.anthony-tong.com/install-mac.sh | bashIt downloads the right build for your Mac, puts Transcribe.app in Applications and opens it.
No password needed. Read the script.
With the stem separator
Add --with-stems to also set up the optional local stem separator (about 0.4 GB, uses the Apple Silicon GPU):
curl -fsSL https://transcribe.anthony-tong.com/install-mac.sh | bash -s -- --with-stemsNot needed for the online MVSEP models: those only need a free API key.
Or download manually
- Apple menu → About This Mac shows which chip you have.
- Double-click the download to unpack
Transcribe.appand drag it into Applications. - In Terminal, run these two lines, or macOS will call the app “damaged”:
xattr -cr /Applications/Transcribe.app codesign --force --deep --sign - /Applications/Transcribe.app
Windows
Windows 10 and 11 · 64-bit
Download the installer, or a portable .exe.
- Run
Transcribe-Setup-1.9.1.exe. - If “Windows protected your PC” appears, click More info, then Run anyway.
- Pick a folder (the default is fine). On the next page, tick Download and set up the stem separator if you want the local one, then click Install.
Portable version
A single .exe that runs from anywhere without installing, e.g. a USB stick. It tells you when a new version is out but can't update itself.
Linux
x86_64 · most current distributions
One line in a terminal for any distribution, or a .deb package.
Any distribution: paste this in a terminal (no sudo; it keeps itself up to date):
curl -fsSL https://transcribe.anthony-tong.com/install-linux.sh | bashInstalls the AppImage in ~/Applications with a menu entry and a transcribe command. Read the script.
Ubuntu, Debian, Mint: the .deb package
Download .deb100 MBOpen it with your software installer, or in the download folder run:
sudo apt install ./Transcribe-1.9.1-linux-amd64.debTranscribe tells you when a new version is out.
With the stem separator, the plain AppImage, uninstalling
Add --with-stems to also set up the optional local stem separator (it uses an NVIDIA GPU with CUDA if there is one):
curl -fsSL https://transcribe.anthony-tong.com/install-linux.sh | bash -s -- --with-stemsOr download the AppImage itself, make it executable (chmod +x) and run it from anywhere:
On Ubuntu 23.10 and later an AppImage only starts with --no-sandbox (the install script sets that up; the .deb doesn't need it). To uninstall: curl -fsSL https://transcribe.anthony-tong.com/install-linux.sh | bash -s -- --uninstall, or sudo apt remove transcribe for the .deb.
Stems (optional)
Split a song into parts (vocals, drums, bass, piano, guitar, strings…) and switch each one on or off while you transcribe. There are two ways to separate, and you can use both:
Local separator
Free and private: nothing is uploaded. Runs Demucs (4 stems, or 6 with guitar and piano) on the GPU when it can (Metal on Apple Silicon, CUDA on NVIDIA), else on the CPU.
Downloads Python, PyTorch and a model the first time: about 0.4 GB on a Mac, 0.5 GB on Windows or Linux, or 3 GB with NVIDIA GPU support. Set it up during install or any time from Stems → Stem Separator Setup & Diagnostics…, which also shows whether the GPU is in use. Stronger RoFormer models can be downloaded too, but they need a good GPU.
MVSEP, with your free API key
MVSEP runs today's best separation models on its own servers, so you get cleaner stems than Demucs without installing anything or owning a GPU. It also has models for single instruments, which is what makes deep splits work:
- Create a free account at mvsep.com.
- Copy your API key from mvsep.com/user-api.
- In Transcribe, open Stems → Manage Stem Splitter Models… (or Models… in the Stems panel), paste the key under MVSEP and click Save key.
- Pick an MVSEP model in the Stems panel and click Separate.
Free accounts: up to 10 minutes and 100 MB per upload, about 50 uploads a day. Longer songs are split into parts automatically. The key is checked with MVSEP, stored encrypted on your computer and only ever sent to mvsep.com. Your audio is uploaded, separated, downloaded and then deleted from MVSEP's server.
song.mp3└ Demucs 4-stem ├ vocals ├ drums ├ bass └ other └ MVSEP Brass ├ brass └ other
Split stems again. Open the Stems panel (Ctrl/Cmd+3), pick the song or any stem and separate it with any model: chain a general model with single-instrument ones to dig out the part you're transcribing. Every split is kept in the project. To share a project without the size, Export Project Copy saves it with only the stems you pick, or none.
Updates
Once installed, Transcribe keeps itself up to date. It checks shortly after it opens and every few hours, downloads new versions in the background, and asks whether to restart now or install when you next quit. To check by hand: Transcribe → Check for Updates… on a Mac, Help → Check for Updates… on Windows and Linux (the AppImage updates itself; the .deb tells you about a new version).
What's new in 1.9.1
- Open now shows audio, videos and Transcribe projects together (a menu in the dialog narrows it)
- AIFF files open again, and Apple Lossless, CAF, WMA and AC-3 files are converted when they're opened (on Windows and Linux these need ffmpeg)
- Opening a video loads its sound
- Files that aren't audio say so instead of "Could not decode"