{"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/improved-training-for-online-end-to-end","title":"Improved training for online end-to-end speech recognition systems","arxiv_id":"1711.02212","date":"2017-11-06","proceeding":null,"authors":["Suyoun Kim","Michael L. Seltzer","Jinyu Li","Rui Zhao"],"abstract":"Achieving high accuracy with end-to-end speech recognizers requires careful\nparameter initialization prior to training. Otherwise, the networks may fail to\nfind a good local optimum. This is particularly true for online networks, such\nas unidirectional LSTMs. Currently, the best strategy to train such systems is\nto bootstrap the training from a tied-triphone system. However, this is time\nconsuming, and more importantly, is impossible for languages without a\nhigh-quality pronunciation lexicon. In this work, we propose an initialization\nstrategy that uses teacher-student learning to transfer knowledge from a large,\nwell-trained, offline end-to-end speech recognition model to an online\nend-to-end model, eliminating the need for a lexicon or any other linguistic\nresources. We also explore curriculum learning and label smoothing and show how\nthey can be combined with the proposed teacher-student learning for further\nimprovements. We evaluate our methods on a Microsoft Cortana personal assistant\ntask and show that the proposed method results in a 19 % relative improvement\nin word error rate compared to a randomly-initialized baseline system.","url_abs":"http://arxiv.org/abs/1711.02212v2","url_pdf":"http://arxiv.org/pdf/1711.02212v2.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":"improved-training-for-online-end-to-end","repo_url":"https://github.com/vadimkantorov/ctc","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"speech-recognition","task_name":"Speech Recognition"},{"task_slug":"speech-recognition-1","task_name":"speech-recognition"}],"methods":[{"method_slug":"label-smoothing","method_name":"Label Smoothing"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.02212","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}