The fast increase of web services and mobile apps, which collect personal\ndata from users, increases the risk that their privacy may be severely\ncompromised. In particular, the increasing variety of spoken language\ninterfaces and voice assistants empowered by the vertiginous breakthroughs in\nDeep Learning are prompting important concerns in the European Union to\npreserve speech data privacy. For instance, an attacker can record speech from\nusers and impersonate them to get access to systems requiring voice\nidentification. Hacking speaker profiles from users is also possible by means\nof existing technology to extract speaker, linguistic (e.g., dialect) and\nparalinguistic features (e.g., age) from the speech signal. In order to\nmitigate these weaknesses, in this paper, we propose a speaker\nde-identification system based on adversarial training and autoencoders in\norder to suppress speaker, gender, and accent information from speech.\nExperimental results show that combining adversarial learning and autoencoders\nincrease the equal error rate of a speaker verification system while preserving\nthe intelligibility of the anonymized spoken content.\n