{"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/scalable-factorized-hierarchical-variational","title":"Scalable Factorized Hierarchical Variational Autoencoder Training","arxiv_id":"1804.03201","date":"2018-04-09","proceeding":null,"authors":["Wei-Ning Hsu","James Glass"],"abstract":"Deep generative models have achieved great success in unsupervised learning\nwith the ability to capture complex nonlinear relationships between latent\ngenerating factors and observations. Among them, a factorized hierarchical\nvariational autoencoder (FHVAE) is a variational inference-based model that\nformulates a hierarchical generative process for sequential data. Specifically,\nan FHVAE model can learn disentangled and interpretable representations, which\nhave been proven useful for numerous speech applications, such as speaker\nverification, robust speech recognition, and voice conversion. However, as we\nwill elaborate in this paper, the training algorithm proposed in the original\npaper is not scalable to datasets of thousands of hours, which makes this model\nless applicable on a larger scale. After identifying limitations in terms of\nruntime, memory, and hyperparameter optimization, we propose a hierarchical\nsampling training algorithm to address all three issues. Our proposed method is\nevaluated comprehensively on a wide variety of datasets, ranging from 3 to\n1,000 hours and involving different types of generating factors, such as\nrecording conditions and noise types. In addition, we also present a new\nvisualization method for qualitatively evaluating the performance with respect\nto the interpretability and disentanglement. Models trained with our proposed\nalgorithm demonstrate the desired characteristics on all the datasets.","url_abs":"http://arxiv.org/abs/1804.03201v2","url_pdf":"http://arxiv.org/pdf/1804.03201v2.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":"scalable-factorized-hierarchical-variational","repo_url":"https://github.com/wnhsu/ScalableFHVAE","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"scalable-factorized-hierarchical-variational","repo_url":"https://github.com/BurnhamG/PyTorch-ScalableFHVAE","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"disentanglement","task_name":"Disentanglement"},{"task_slug":"hyperparameter-optimization","task_name":"Hyperparameter Optimization"},{"task_slug":"robust-speech-recognition","task_name":"Robust Speech Recognition"},{"task_slug":"speaker-verification","task_name":"Speaker Verification"},{"task_slug":"speech-recognition","task_name":"Speech Recognition"},{"task_slug":"variational-inference","task_name":"Variational Inference"},{"task_slug":"voice-conversion","task_name":"Voice Conversion"},{"task_slug":"speech-recognition-1","task_name":"speech-recognition"}],"methods":[{"method_slug":"interpretability","method_name":"Interpretability"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}