🎤 Vocal Remover (Optimized CPU Edition)

Kim_Vocal_2 · MDX-Net ONNX · Best quality-per-byte vocal separator

Upload any song to isolate vocals and instrumental stems. Supports MP3, WAV, FLAC, M4A, OGG, and more.

⚡ Running on CPU — With Fast Mode enabled, a typical 3-min track takes ~40 seconds.
The model (~67 MB) is downloaded automatically on first run, then cached for the session.

Upload Audio

🎤 Vocals Only

🎵 Instrumental

Waveforms & Spectrograms

📊 Quality Metrics

Model Comparison

Model	Size	Vocal SDR	ONNX
Kim_Vocal_2 ✅	67 MB	~8.9 dB	✅
UVR-MDX-NET-Voc_FT	67 MB	~8.7 dB	✅
htdemucs_ft (Demucs 4)	330 MB	~9.2 dB	❌
BS-Roformer	430 MB	~12.9 dB	❌
Kim_Vocal_2 is the best quality-per-byte ONNX vocal model — ideal for free-tier CPU inference.