{"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/supervised-and-unsupervised-speech","title":"Supervised and Unsupervised Speech Enhancement Using Nonnegative Matrix Factorization","arxiv_id":"1709.05362","date":"2017-09-15","proceeding":null,"authors":["Nasser Mohammadiha","Paris Smaragdis","Arne Leijon"],"abstract":"Reducing the interference noise in a monaural noisy speech signal has been a\nchallenging task for many years. Compared to traditional unsupervised speech\nenhancement methods, e.g., Wiener filtering, supervised approaches, such as\nalgorithms based on hidden Markov models (HMM), lead to higher-quality enhanced\nspeech signals. However, the main practical difficulty of these approaches is\nthat for each noise type a model is required to be trained a priori. In this\npaper, we investigate a new class of supervised speech denoising algorithms\nusing nonnegative matrix factorization (NMF). We propose a novel speech\nenhancement method that is based on a Bayesian formulation of NMF (BNMF). To\ncircumvent the mismatch problem between the training and testing stages, we\npropose two solutions. First, we use an HMM in combination with BNMF (BNMF-HMM)\nto derive a minimum mean square error (MMSE) estimator for the speech signal\nwith no information about the underlying noise type. Second, we suggest a\nscheme to learn the required noise BNMF model online, which is then used to\ndevelop an unsupervised speech enhancement system. Extensive experiments are\ncarried out to investigate the performance of the proposed methods under\ndifferent conditions. Moreover, we compare the performance of the developed\nalgorithms with state-of-the-art speech enhancement schemes using various\nobjective measures. Our simulations show that the proposed BNMF-based methods\noutperform the competing algorithms substantially.","url_abs":"http://arxiv.org/abs/1709.05362v1","url_pdf":"http://arxiv.org/pdf/1709.05362v1.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":"supervised-and-unsupervised-speech","repo_url":"https://github.com/mohammadiha/bnmf","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"denoising","task_name":"Denoising"},{"task_slug":"speech-denoising","task_name":"Speech Denoising"},{"task_slug":"speech-enhancement","task_name":"Speech Enhancement"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1709.05362","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}