Text to Speech AI Agents Music Speech to Text Voice Cloning Voice Design Marketplace More
Report Bug / Feature Request

AI Stem Splitter

Split any song into individual stems: vocals, drums, bass, and other instruments. Powered by state-of-the-art AI for music production, remixing, and practice.

Sign Up Free

Upload Music Track

3,000-5,000 characters per song

Drag & drop your file here, or browse

Supports MP3, WAV, FLAC, OGG, M4A. Max 500 MB (2 GB on paid plans). Songs up to 10 minutes.

file.mp3

0 MB

Stem Splitting Settings

Vocals from a Mel-Band RoFormer model, drums, bass and other from an SCNet model, lossless WAV output, ready for your DAW.

Free 2-stem splitting — no account needed.

How It Works

Upload Your Track
AI Separates Each Stem
Preview & Download Stems

Splitting your track into stems...

Using Mel-Band RoFormer + SCNet for 4 Stems
Taking a while? Your result will appear in your generation history when ready.
Stem Splitting Complete

Vocals

0:00

Drums

0:00

Bass

0:00

Other

0:00
Stems Created 4
Song Duration 4:12
Processing Time 18.3s

What Are Audio Stems?

Audio stems are the individual instrument or vocal tracks that make up a song. Professional music is typically mixed from dozens of individual tracks, but stem splitting uses AI to reverse-engineer the mix and extract the primary components.

Vocals Singing, rapping, harmonies, backing vocals, ad-libs
Drums Kick, snare, hi-hats, cymbals, toms, percussion
Bass Bass guitar, synth bass, sub bass, 808s
Other Everything else: guitars, piano, keys, strings, brass, synths, effects

Model Details

Vocals Mel-Band RoFormer

Separates the vocals in every split. A transformer attends across time and across mel-scaled frequency bands of the spectrogram, and the instrumental is the mix with those vocals removed. Measured on 50 MUSDB18 test excerpts: 11.3 dB vocal SDR and 16.0 dB instrumental SDR.

Stems: 2, 4

Drums SCNet

Splits the instrumental into drums, bass and other instruments in 4-stem mode. It divides the spectrogram into frequency sub-bands and gives the low bands, where bass and kick drums sit, more detail.

Stems: 4

Tips for Best Results

  • Use lossless input (WAV/FLAC) for the cleanest separation
  • 4 stems is the sweet spot for most music production needs
  • Guitar, piano and synths all land in the Other stem
  • Songs with clear arrangements (less layering) produce better results
  • Export as WAV if you plan to import into a DAW for editing
  • For vocal-only removal, try our specialized Vocal Remover

How It Works

Our deep learning models analyze the spectral and temporal patterns in your music to isolate individual instrument groups with remarkable precision.

Step 1

Upload Your Track

Drag and drop your music file in any common audio format. We accept MP3, WAV, FLAC, OGG, and M4A files up to 50MB. For best results, use lossless formats like WAV or FLAC. Your file is securely processed on our GPU servers and automatically deleted within 1 hour.

Step 2

AI Separates Each Stem

A Mel-Band RoFormer model first separates the vocals, using transformer attention across time and mel-scaled frequency bands of the spectrogram. For 4 stems, an SCNet model then splits the remaining instrumental into drums, bass and other instruments.

Step 3

Preview & Download Stems

Each extracted stem gets its own audio player so you can preview individual instruments in isolation. Solo or mute specific stems to hear how they interact. Download each stem separately or grab all of them in a convenient ZIP archive ready to import into your DAW.

Stem Splitting Use Cases

AI-powered stem splitting is revolutionizing music production, education, and creative workflows across the industry.

Music Production & Remixing

Extract individual stems to create remixes, mashups, and new arrangements. Sample drum patterns, bass lines, or vocal hooks from existing recordings. Layer stems from different songs to create unique productions. Essential for producers working with reference tracks or creating official remixes.

DJ Sets & Performance

Prepare stems for live DJ sets and performances. Create acapella drops, isolated drum breaks, and bass-only versions for creative blending. Build transition tracks that seamlessly connect songs. Mix vocal stems from one track over the instrumental of another for unique live mashups.

Practice & Study

Isolate specific instrument parts to practice along with. Drummers can mute the drum track and play along with the rest of the band. Bassists can isolate the bass line to study technique. Guitarists and keyboard players can solo the Other stem to hear their parts without vocals, drums or bass.

Music Education

Break down songs into components for music theory analysis. Help students understand how instruments interact in a mix. Demonstrate arrangement techniques by showing how each stem contributes to the final song. Create custom practice tracks for music classes.

Sampling & Sound Design

Extract clean samples from any recording for use in new productions. Isolate one-shot drum hits, bass tones, vocal chops, and musical phrases. Build sample packs and sound libraries from existing recordings. Create unique textures by processing individual stems through effects chains.

Film & Media Post-Production

Extract music stems for film and TV post-production. Adjust the balance between vocals and instruments in soundtrack mixes. Create instrumental versions of licensed music for background use. Isolate specific elements for sound design and Foley work in video production.

Karaoke & Cover Versions

Create perfect karaoke backing tracks by removing vocals. Record your own vocals over the original instrumental arrangement. Use drum and bass stems as a rhythm section backing track for acoustic performances or band rehearsals. Build custom backing tracks for live shows.

Audio Analysis & Transcription

Isolate stems to create accurate musical transcriptions. Hear individual instrument parts clearly for notation purposes. Analyze mixing techniques by studying how each stem is processed. Perfect for musicologists, transcription services, and arrangement analysis.

The Most Powerful Free Stem Splitter Online

Two Specialized Models

Each part of the split uses a model trained for that job. Vocals come from a Mel-Band RoFormer model that measured 11.3 dB vocal SDR on 50 MUSDB18 test excerpts, about 3 dB above our previous engine. In 4-stem mode an SCNet model separates drums, bass and other instruments from the instrumental. Guitar, piano and synths stay together in the Other stem.

Free 2-Stem Splitting

Get started for free with 2-stem separation. Split any song into vocals and instrumental without creating an account or spending characters. The free split uses the same vocal model as account splits, with no watermarks. Create a free account to split drums, bass and other instruments in 4-stem mode.

GPU-Accelerated Processing

Stem splitting is computationally intensive. TTS.ai runs the separation models on dedicated NVIDIA GPUs. Processing time grows with song length, and 4 stems take longer than 2 because a second model runs on the instrumental.

REST API for Batch Processing

Process entire music libraries programmatically with our developer API. Build stem splitting into your own applications, music production tools, or DJ preparation workflows. The API supports 2 and 4 stems. Its model parameter still accepts demucs and spleeter for compatibility, and both run the same engine. API access included on every plan with Python, JavaScript, and cURL SDKs.

Stem Splitting Plans

Start free, upgrade when you need more

Free
  • 2-stem split (Vocals + Instrumental)
  • Mel-Band RoFormer vocal model
  • 10-minute songs
  • No account required
Most Popular
Free Account
  • 4-stem split (Vocals, Drums, Bass, Other)
  • SCNet drums, bass and other
  • Per-stem download + ZIP
  • 15,000 free characters
Sign Up Free
Pro
  • Uploads up to 2 GB
  • Longer audio files
  • Batch processing
  • API access
Upgrade

Frequently Asked Questions

Stem splitting separates a mixed audio track into individual components (stems): vocals, drums, bass, and other instruments. This is invaluable for music production, remixing, sampling, transcription, and karaoke.

You can split a track into 2 stems (vocals and instrumental) or 4 stems (vocals, drums, bass, and other instruments). Guitar, piano, synths and everything else that is not vocals, drums or bass stay together in the Other stem.

Vocals are separated by a Mel-Band RoFormer model in every split. In 4-stem mode, an SCNet model then separates the drums, bass and other instruments from the instrumental. There is one engine for every tier, so the free 2-stem split uses the same vocal model as account splits.

On 50 MUSDB18 test excerpts the vocal model measured 11.3 dB vocal SDR and 16.0 dB instrumental SDR (signal-to-distortion ratio, higher is cleaner), about 3 dB better on both than the engine we used before. Some bleed-through between stems is normal, especially on heavily layered or compressed mixes.

Processing time depends on the track length and the number of stems. A 4-stem split takes longer than a 2-stem split, because a second model separates the instrumental after the vocals are removed.

Stems you separate on any paid plan can be used commercially, and upgrading also covers stems you made earlier on the free tier; free-tier output is for personal, non-commercial use only. Both separation models are released under licenses that allow commercial use: MIT for the Mel-Band RoFormer vocal model and Apache 2.0 for the SCNet model. In every case you must have the rights to the original audio: separating copyrighted music does not grant you new rights to the individual stems.

Batch processing is available via our API, allowing you to submit multiple tracks for stem separation in an automated workflow. The web interface handles one track at a time with instant preview and download of all stems.

Not as its own stem. The 4-stem mode puts guitars, piano, synths, keyboards and every other instrument together in the Other stem. A dedicated guitar or piano stem is not available.

Yes, each stem is exported as a separate audio file that can be imported into any DAW (Ableton, Logic Pro, FL Studio, etc.) for remixing. Adjust levels, add effects, or replace individual stems to create new arrangements.

Stem splitting uses 2,000 characters per track processed. Free accounts receive 15,000 characters on signup. All paid plans include generous character allowances for regular use.

We accept MP3, WAV, FLAC, OGG, M4A, and WEBM files up to 50MB. For best separation quality, upload lossless formats (WAV, FLAC). Output stems are provided in WAV format to preserve maximum quality.

Our vocals come from a Mel-Band RoFormer model, a transformer architecture for music source separation, and the 4-stem split adds an SCNet model for drums, bass and other instruments. Rather than a ranking, we publish what we measured: 11.3 dB vocal SDR and 16.0 dB instrumental SDR on 50 MUSDB18 test excerpts.
5.0/5 (1)

What could we improve? Your feedback helps us fix issues.

Split Any Song into Stems for Free

Join thousands of musicians and producers using TTS.ai. Free 2-stem splitting available now. Get 15,000 characters with a free account for 4-stem mode.

Sign Up Free View Pricing