{"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/sampling-strategies-in-siamese-networks-for","title":"Sampling strategies in Siamese Networks for unsupervised speech representation learning","arxiv_id":"1804.11297","date":"2018-04-30","proceeding":null,"authors":["Rachid Riad","Corentin Dancette","Julien Karadayi","Neil Zeghidour","Thomas Schatz","Emmanuel Dupoux"],"abstract":"Recent studies have investigated siamese network architectures for learning\ninvariant speech representations using same-different side information at the\nword level. Here we investigate systematically an often ignored component of\nsiamese networks: the sampling procedure (how pairs of same vs. different\ntokens are selected). We show that sampling strategies taking into account\nZipf's Law, the distribution of speakers and the proportions of same and\ndifferent pairs of words significantly impact the performance of the network.\nIn particular, we show that word frequency compression improves learning across\na large range of variations in number of training pairs. This effect does not\napply to the same extent to the fully unsupervised setting, where the pairs of\nsame-different words are obtained by spoken term discovery. We apply these\nresults to pairs of words discovered using an unsupervised algorithm and show\nan improvement on state-of-the-art in unsupervised representation learning\nusing siamese networks.","url_abs":"http://arxiv.org/abs/1804.11297v2","url_pdf":"http://arxiv.org/pdf/1804.11297v2.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":"sampling-strategies-in-siamese-networks-for","repo_url":"https://github.com/Rachine/sampling_siamese2018","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"GPL-3.0"}},{"paper_slug":"sampling-strategies-in-siamese-networks-for","repo_url":"https://github.com/bootphon/abnet3","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":"representation-learning","task_name":"Representation Learning"},{"task_slug":"speech-representation-learning","task_name":"Speech Representation Learning"}],"methods":[{"method_slug":"siamese-network","method_name":"Siamese Network"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1804.11297","atlas_url":"https://app.syntology.ai/?focus=1804.11297","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}