Home

I’m a researcher at DFKI and a third-year PhD student at TU Berlin. My research focuses on speaker anonymization, i.e. how can speaker information be concealed in a speech signal, while preserving its utility. How utility is defined depends on the use case. For example, at DFKI we work with therapists from Charité University to develop a real-time anonymizer that can be used for therapy. This use case requires the anonymized speech to be expressive, for therapists to fully understand what their patients convey to them.

In the past I’ve also worked on accented speech recognition and NLP topics like knowledge graphs and subject indexing.

My posts

  • Introduction to speaker anonymization

    Yesterday I gave a lecture on speaker anonymization at the biometrics seminar at TU Berlin. My aim was to give a broad introduction to the field, explaining the core ideas and challenges, and providing relevant references. I was allowed to record it; you can find a link to the video below, as well as the link to a compact version of the slides. Hopefully, newcomers to the field find this overview helpful to get started. If you find errors of any kind, please let me know and I will update the material.

  • Private kNN-VC

    Private kNN-VC is an anonymizer for speech. Given a speech sample and a target speaker, it outputs speech that preserves the content of the input speech but sounds like the target speaker. To enhance privacy, the duration and variation of the phones are anonymized, as they contain speaker information.

subscribe via RSS