{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/monaural-audio-speaker-separation-with-source","title":"Monaural Audio Speaker Separation with Source Contrastive Estimation","arxiv_id":"1705.04662","date":"2017-05-12","proceeding":null,"authors":["Cory Stephenson","Patrick Callier","Abhinav Ganesh","Karl Ni"],"abstract":"We propose an algorithm to separate simultaneously speaking persons from each\nother, the \"cocktail party problem\", using a single microphone. Our approach\ninvolves a deep recurrent neural networks regression to a vector space that is\ndescriptive of independent speakers. Such a vector space can embed empirically\ndetermined speaker characteristics and is optimized by distinguishing between\nspeaker masks. We call this technique source-contrastive estimation. The\nmethodology is inspired by negative sampling, which has seen success in natural\nlanguage processing, where an embedding is learned by correlating and\nde-correlating a given input vector with output weights. Although the matrix\ndetermined by the output weights is dependent on a set of known speakers, we\nonly use the input vectors during inference. Doing so will ensure that source\nseparation is explicitly speaker-independent. Our approach is similar to recent\ndeep neural network clustering and permutation-invariant training research; we\nuse weighted spectral features and masks to augment individual speaker\nfrequencies while filtering out other speakers. We avoid, however, the severe\ncomputational burden of other approaches with our technique. Furthermore, by\ntraining a vector space rather than combinations of different speakers or\ndifferences thereof, we avoid the so-called permutation problem during\ntraining. Our algorithm offers an intuitive, computationally efficient response\nto the cocktail party problem, and most importantly boasts better empirical\nperformance than other current techniques.","url_abs":"http://arxiv.org/abs/1705.04662v1","url_pdf":"http://arxiv.org/pdf/1705.04662v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"monaural-audio-speaker-separation-with-source","repo_url":"https://github.com/lab41/magnolia","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"descriptive","task_name":"Descriptive"},{"task_slug":"speaker-separation","task_name":"Speaker Separation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}