Demucs
Demucs separates audio into clean vocals, drums, bass, and other stems.
A tradução deste card ainda não está disponível; parte do conteúdo aparece em inglês.
Demucs excels at separating complex audio mixtures into distinct stems like vocals, drums, bass, and other instruments using deep learning. It works well on full songs, podcasts, or live recordings, preserving audio quality while minimizing artifacts. Users can isolate a singer’s voice for karaoke, extract drum loops for beats, or mute instruments to practice along.
Use Demucs when you need precise, high-fidelity separation — whether you're a musician creating backing tracks, a podcaster cleaning up dialogue, or a content creator sampling audio. Unlike older methods, it handles overlapping frequencies and reverb with remarkable clarity.
What sets Demucs apart is its balance of speed and accuracy: it delivers professional-grade stem separation without requiring studio tools or manual editing. The model adapts to various genres and mixing styles, making it reliable for both casual and professional creative workflows.
- Isolate vocals for karaoke or remixing
- Extract drum or bass loops for beatmaking
- Remove instruments to practice playing along
- Clean up dialogue in podcasts or interviews
- Sample specific stems for music production
Produces clean stems with minimal distortion or bleeding between instruments.
Effective on pop, rock, hip-hop, classical, and mixed-genre audio.
Maintains original tone and dynamics better than many traditional methods.
- Use high-quality source audio (320kbps MP3 or WAV) for best results.
- Trim silence at the start/end to reduce processing time.
- Try different output formats to find the right balance of quality and file size.
- Listen to each stem individually before combining them in a mix.
- Struggles with very low-quality or heavily compressed audio.
- May misseparate instruments with similar frequency ranges (e.g., synth and pad).
- Processing time increases with longer audio files.
Isolate only the vocals from a classic soul recording to create a pure acappella version.
Separate a hip-hop track into vocals, kick, snare, hi-hat, bass, and synth for full creative control.
Extract strings and brass from a movie soundtrack to study orchestration layer by layer.
Controls randomness in processing; using the same seed gives repeatable results.
Number of refinement steps; more steps can improve quality but increase processing time.
The length of audio to process — longer files take more time but aren’t limited by the model.
Adjusts how strictly the model follows the separation — higher values may reduce artifacts but could thin out sound.
Os preços exibidos são o que você paga na Muitto. Sem surpresas.
Can Demucs separate live recordings?
Yes, it works well on live audio as long as the mix isn’t overly noisy or distorted.
Does it work on mono audio?
It processes mono tracks, but stereo separation is more effective due to spatial cues.
Is there a limit on file length?
There’s no hard limit, but longer files take more time and memory to process.