Project TitleREPET (REpeating Pattern Extraction Technique)
Track Code2011-063
Short Description

REPET uses a simple but reliable technique for music/voice separation. #software #soundsynthesis #datamining


Currently available music separation methods are more demanding and complex, demanding system “training,” user designation of special audio features, and extensive processing time to support their complex frameworks.  REPET demands none of this, and is simple, fast and completely automatable.  Evaluation of 1,000 song clips showed that REPET achieves better separation performance than existing automatic approaches, and in much simpler fashion.  

REPET exploits core principle in music – repetition. This especially applies to popular songs.  REPET separates music from voice, simply by extracting the repeating musical structure.  It first finds the period of the repeating structure. Then, the spectrogram is segmented at period boundaries and the segments are averaged to create a repeating segment model. Finally, each time-frequency bin in a segment is compared to the model, and the mixture is partitioned using binary time-frequency masking by labeling bins similar to the model as the repeating background. Because REPET utilizes “self-similarity” between repeating segments, it works on a variety of audio signals having one or more repeating patterns within a recording. 

TagsSOFTWARE: data mining, SOFTWARE: sound synthesis
Posted DateOct 5, 2011 5:56 PM


  • Entirely automated – no user input needed
  • No system “training” needed
  • Simple processing – no complex frameworks needed


  • Music search engine by voice recognition
  • Music/Voice transcription
  • Audio post-production
  • Audio analysis for remixing
  • Sound separation
  • Sample-based musical composition
  • Karaoke: Active noise removal


Northwestern seeks commercial partners.  A patent application has been filed. 

Contact Information

Arjan Quist, PhD 
Invention Manager