Splittr – AI Stem Separator

About

Splittr uses AI to separate any YouTube video into individual stems — vocals, drums, bass, and other instruments — all processed locally on your device.

FEATURES

4-Stem AI Separation: Powered by Meta's HTDemucs neural network, Splittr isolates vocals, drums, bass, and other instruments with high accuracy.

100% Local Processing: All AI inference runs on your device using WebGPU (GPU-accelerated) or WASM (CPU fallback). No audio is ever uploaded to any server.

Automatic Key Detection: Identifies the musical key of any track with Camelot notation for easy mixing reference.

Automatic BPM Detection: Accurate tempo detection using multi-band spectral flux analysis with perceptual weighting.

Real-Time Transpose: Pitch-shift playback up or down in semitones without affecting tempo.

BPM Adjustment: Speed up or slow down playback while maintaining pitch.

Stem Toggle: Mute or solo any combination of vocals, drums, bass, and other instruments.

HOW IT WORKS

Navigate to any YouTube video
The Splittr button appears in the bottom-right corner
Click "Separate Stems" to start AI processing
On first use, a 172 MB AI model is downloaded and cached locally
Processing takes 30-90 seconds depending on your hardware
Toggle individual stems on/off and adjust pitch or tempo freely


REQUIREMENTS

Chrome browser with WebGPU support (Chrome 113+) for GPU acceleration
Falls back to WASM (CPU) on older hardware
~200 MB of storage for the AI model (downloaded once, cached permanently)

PRIVACY

Your audio never leaves your device. Splittr performs all AI processing locally using your GPU or CPU. The only network request is the one-time AI model download from Hugging Face. No analytics, no tracking, no data collection.