facebookresearch/demucsArchived

Features

• Source Separation Tools - Provides software tools for splitting mixed audio recordings into individual instrument and vocal tracks.
• Audio Source Separation Models - Uses deep learning models to decompose mixed music tracks into individual stems like drums and vocals.
• Hybrid Spectral-Waveform Separators - Combines spectral and waveform analysis for high-fidelity source separation.
• Audio De-mixing Software - Automates the creation of acapellas and backing tracks by removing specific instruments from a mix.
• Hybrid Domain Audio Processing - Combines spectral and waveform analysis to capture both fine-grained transients and overall spectral shape.
• Audio Stem Extractors - Isolates a single audio source from a mixed recording to create standalone acapellas or backing tracks.
• Vocal Isolation - Removes all instruments from a song to isolate the vocal performance for remixes or sampling.
• Vocal-to-Instrumental Converters - Creates instrumental versions of songs by removing the vocal tracks.
• Convolutional Neural Networks - Implements convolutional neural networks to identify and isolate spatial patterns within audio spectrograms.
• U-Net Architectures - Employs a U-Net architecture to restore high-resolution audio details using encoder skip-connections.
• Time-Frequency Mapping - Uses Short-Time Fourier Transforms to transform raw waveforms into time-frequency representations for efficient separation.
• Time-Domain Signal Refinement - Applies final processing directly to the time-domain signal to correct artifacts introduced by spectral processing.

Star history