What is Stem Separation?
Stem separation (also called source separation or demixing) is an AI powered audio processing technique that decomposes a mixed audio recording into its constituent parts, typically drums, bass, vocals, melody, lead instruments, and sound effects. The AI analyses the frequency spectrum and temporal patterns to isolate each source, transforming a single mixed file into multiple independent stems.
How AI Stem Separation Works
Traditional audio mixing is additive: you record separate tracks and combine them. Stem separation is subtractive: you take a finished mix and the AI estimates which parts belong to each instrument. It works by analyzing the frequency spectrum, drums occupy specific frequency bands, vocals sit in the mid range, bass in the low end, and separating signals based on their spectral characteristics.
Modern stem separation uses deep neural networks trained on large datasets of isolated instrument tracks alongside their mixed versions. The model learns to predict the contribution of each instrument to each time frequency bin of the mixed signal.
How Creators Use Stem Separation
The most common use case is removing vocals to create an instrumental version of a song. But stem separation goes further: isolating the drums for a transition effect, muting the melody to record a voiceover, boosting the bass for a cinematic moment, or extracting a clean vocal for a remix.
Tonr's Pro plan includes 6 stem AI separation on every generated track and any uploaded audio. Each stem outputs as a separate WAV file for your DAW or editor.
stem separation / source separation / demixing / audio AI / music production