Stem separation — also called source separation — is the process of breaking a finished, mixed song into its individual audio components. Think of it as unmixing a smoothie back into its individual fruits.
What Is a "Stem" in Music?
In music production, a stem is a sub-mix of related tracks exported as a single audio file. A typical set of stems from a pop song might include:
- Vocals — lead vocal, harmonies, backing vocals
- Drums — kick, snare, hi-hats, cymbals
- Bass — bass guitar or synth bass
- Other — guitars, keys, synths, strings, everything else
When a producer finishes a song, they mix these stems together into a single stereo file — the song you hear on Spotify. Stem separation tries to reverse this process.
How Does AI Stem Separation Work?
Method 1: Signal Processing (Filter Banks)
This approach uses the known frequency ranges of instruments to separate them:
- Bass typically lives below 300 Hz → apply a low-pass filter
- Drums have transient energy at 60–150 Hz (kick/snare) and 3 kHz+ (hi-hats)
- Vocals sit in the 300 Hz – 3 kHz range
This is what 7By.in's Stem Splitter uses. It runs in your browser instantly with no upload needed.
Method 2: Deep Learning (Neural Networks)
Systems like Meta's Demucs and Deezer's Spleeter are trained on thousands of songs where the original stems are known. The model learns to predict which frequencies belong to which source. These produce higher quality results but require significant GPU compute — usually a server.
2-Stem vs 4-Stem Separation
What Are Stems Used For?
- Remixing — use isolated drums or bass from a song in your own production
- Music learning — isolate a guitar riff to learn it by ear
- Karaoke — remove vocals to sing along
- Podcasting — extract a specific instrument as background music
- Film scoring — study how arrangements work in finished recordings
Limitations of Stem Separation
No tool can perfectly separate all songs. Common challenges:
- Instruments that share frequency ranges bleed into each other
- Heavy reverb and compression make separation harder
- Mono recordings have no stereo information to exploit
- Dense arrangements are harder than sparse ones
Try Stem Splitting Free
Two free splits per day. Runs entirely in your browser.
Open Stem Splitter →