{"language":[{"iso":"eng"}],"file":[{"relation":"main_file","file_size":272390,"creator":"frra","date_created":"2023-10-20T08:20:58Z","content_type":"application/pdf","access_level":"closed","file_id":"48359","success":1,"file_name":"arxiv.pdf","date_updated":"2023-10-20T08:20:58Z"}],"date_updated":"2023-11-22T13:44:33Z","title":"On Feature Importance and Interpretability of Speaker Representations","conference":{"location":"Aachen","end_date":"2023-09-22","name":"ITG Conference on Speech Communication","start_date":"2023-09-20"},"publication":"ITG Conference on Speech Communication","abstract":[{"lang":"eng","text":"Unsupervised speech disentanglement aims at separating fast varying from\r\nslowly varying components of a speech signal. In this contribution, we take a\r\ncloser look at the embedding vector representing the slowly varying signal\r\ncomponents, commonly named the speaker embedding vector. We ask, which\r\nproperties of a speaker's voice are captured and investigate to which extent do\r\nindividual embedding vector components sign responsible for them, using the\r\nconcept of Shapley values. Our findings show that certain speaker-specific\r\nacoustic-phonetic properties can be fairly well predicted from the speaker\r\nembedding, while the investigated more abstract voice quality features cannot."}],"date_created":"2023-10-20T08:04:46Z","external_id":{"arxiv":["2310.12599"]},"_id":"48355","file_date_updated":"2023-10-20T08:20:58Z","status":"public","year":"2023","author":[{"id":"72602","first_name":"Frederik","last_name":"Rautenberg","full_name":"Rautenberg, Frederik"},{"first_name":"Michael","id":"49871","last_name":"Kuhlmann","full_name":"Kuhlmann, Michael"},{"first_name":"Jana","last_name":"Wiechmann","full_name":"Wiechmann, Jana"},{"first_name":"Fritz","full_name":"Seebauer, Fritz","last_name":"Seebauer"},{"first_name":"Petra","full_name":"Wagner, Petra","last_name":"Wagner"},{"first_name":"Reinhold","id":"242","last_name":"Haeb-Umbach","full_name":"Haeb-Umbach, Reinhold"}],"user_id":"72602","main_file_link":[{"url":"https://arxiv.org/abs/2310.12599","open_access":"1"}],"oa":"1","type":"conference","citation":{"ieee":"F. Rautenberg, M. Kuhlmann, J. Wiechmann, F. Seebauer, P. Wagner, and R. Haeb-Umbach, “On Feature Importance and Interpretability of Speaker Representations,” presented at the ITG Conference on Speech Communication, Aachen, 2023.","bibtex":"@inproceedings{Rautenberg_Kuhlmann_Wiechmann_Seebauer_Wagner_Haeb-Umbach_2023, title={On Feature Importance and Interpretability of Speaker Representations}, booktitle={ITG Conference on Speech Communication}, author={Rautenberg, Frederik and Kuhlmann, Michael and Wiechmann, Jana and Seebauer, Fritz and Wagner, Petra and Haeb-Umbach, Reinhold}, year={2023} }","ama":"Rautenberg F, Kuhlmann M, Wiechmann J, Seebauer F, Wagner P, Haeb-Umbach R. On Feature Importance and Interpretability of Speaker Representations. In: <i>ITG Conference on Speech Communication</i>. ; 2023.","short":"F. Rautenberg, M. Kuhlmann, J. Wiechmann, F. Seebauer, P. Wagner, R. Haeb-Umbach, in: ITG Conference on Speech Communication, 2023.","apa":"Rautenberg, F., Kuhlmann, M., Wiechmann, J., Seebauer, F., Wagner, P., &#38; Haeb-Umbach, R. (2023). On Feature Importance and Interpretability of Speaker Representations. <i>ITG Conference on Speech Communication</i>. ITG Conference on Speech Communication, Aachen.","chicago":"Rautenberg, Frederik, Michael Kuhlmann, Jana Wiechmann, Fritz Seebauer, Petra Wagner, and Reinhold Haeb-Umbach. “On Feature Importance and Interpretability of Speaker Representations.” In <i>ITG Conference on Speech Communication</i>, 2023.","mla":"Rautenberg, Frederik, et al. “On Feature Importance and Interpretability of Speaker Representations.” <i>ITG Conference on Speech Communication</i>, 2023."},"project":[{"_id":"129","name":"TRR 318 - C06: TRR 318 - Technisch unterstütztes Erklären von Stimmcharakteristika (Teilprojekt C06)","grant_number":"438445824"}],"ddc":["000"],"has_accepted_license":"1","department":[{"_id":"54"},{"_id":"660"}]}