Xtrax Stems is a Mac application that creates separate stems from a full mix. The input is any audio file, and the output is three WAV files labeled Vocals, Drums, and Music (everything but drums and vocals). Once separated, you can export the track in Native Instruments Stems file format.
Processing occurs offline, not in your computer, so an Internet connection is required. Importing a song immediately uploads it to Audionamix’s cloud servers. When extraction is complete, the application downloads the stems.
At the top of Xtrax Stems’ GUI are three transport buttons—Play, Return to Beginning, and Loop—and a ruler-like timeline. Below is a Source button for playing the original file, level slider and bar graph meter, a pan slider, Mute and Solo buttons, and a waveform display. At the bottom are four buttons for selecting the separation algorithm, a bar indicating upload and extraction progress, and an overall level slider and meter. Clicking in the timeline repositions the playhead, and dragging in the timeline selects a region to loop.
To import an audio file, click and drag it into the window or choose Import from the File menu. The software recognizes WAV, AIFF, MP3, AAC, and M4A files up to 32 bits, 96 kHz. When you click on one of four separation algorithm buttons—Automatic, Automatic HQ, Generic, or Generic HQ—you’re prompted to log in. The file uploads to the server, and extraction begins. Uploading and extraction each take approximately the length of the song, but HQ separations take longer.
Once extracted, the three stems are displayed, each with its level, pan, mute, and solo controls. You can solo each or play them together in any combination, and you can alter their mix by changing individual levels and pan positions. After running all four algorithms, clicking on any separation algorithm button loads its extraction. Switching from one to another pauses everything for about five seconds.
I extracted stems from several types of songs with female or male voices and simple or complex arrangements. For songs with vocals, I usually started with the Automatic algorithm, which uses a process called Automatic Voice Activity Detection. If the resulting Vocals track had dropouts, I tried Generic, which is recommended for enhancing instrumental solos.
In practice, I couldn’t tell that any algorithm was consistently better than another for any particular type of audio. On some songs, Automatic yielded the clearest vocals; on others, Generic was better. I had to try them all and choose the best.
On one song with female vocals, no matter which algorithm I chose, the flute and soprano sax were clearer than the voice on the Vocals stem, and hand percussion was clearer on the Music track than on Drums. On another song, Generic made the voice sound rather saxophone-like, and snippets of bass were on all three stems. To some degree, every part bled into stems where they didn’t belong, though usually not enough to be a problem.
I’d hoped Xtrax Stems would produce stems I could use for mixdowns, but I found it more useful to minimize vocals or clarify instrumental parts to make them easiy to learn. However minor, unwanted artifacts are all but impossible to eliminate, and they’re most audible when you solo a stem. Yet, the Xtrax Stems technology seems miraculous, and I hope Audionamix keeps refining it.
Usually distinguishes vocals, drums, and solo instruments from other parts. Supports all common audio formats. Easy to use.
Results seldom sound completely natural. Running all four algorithms takes a while.
Writer, synthesist, and Electronic Musician editor-at-large Geary Yelton lives in Asheville, North Carolina.