Idioma: English (mozilla)
The session takes place in SPATIAL CHAT -> Session Specific Rooms -> Fair Voice
https://spatial.chat/s/mozilla-festival-2022?guest&room=8YoigrnYdDGinEHdU5Zp
Your voice contains a lot of personal information: your identity, age, sex, health, stress, emotion, ethnicity and intoxication are just some of the things that can be deduced from your speech signal, without even knowing the content of what you said. Speaker verification is a voice-based biometric technology that identifies speakers from their voice. It is used in voice assistants, call centers, and forensics, and relies on vast amounts of personal data to train advanced machine learning algorithms. There are many parallels between the history and technologies of face recognition and speaker verification. However, while bias and discrimination are recognised as significant challenges in face recognition, little attention has been paid to them in speaker verification. This is a problem, as speaker verification systems are increasingly used in sensitive domains (e.g., financial services, pensioner proof-of-life verification, health and elderly care).
In this session we host a dialogue to introduce speaker verification and examine what is at stake when the technology is deployed at scale for voice-based authentication. We have invited panelists with diverse expertise in speech science, algorithmic activism and fairness, media and art to guide us through an interactive session with MozFest participants. The session is facilitated by the Fair EVA project of the Trustworthy AI working group. We will share what Fair EVA is doing to combat biased voice biometrics, and what you can do to contribute. Join live or watch the session later!
Panelists: Halsey Burgund, Johann Diedrick, Kathleen Siminyu, Wiebke Toussaint
The goal of our session is to introduce the audience to their voice as a prevalent, yet hidden form of biometric identification. We want to raise public awareness about the extent of personal identifiable information contained in voice data. Together with panelists and the audience we explore what is at stake when voice biometrics are deployed at scale. Our intended outcome is an engaging discussion and an audience that is better informed about speaker verification technologies. We also aim to use the session as a call to action for volunteers to join the Fair EVA project and assist us on two tasks:
1) expanding the Fair EVA Speaker Verification Knowledge Database, a comprehensive guide to speaker verification technology vendors, applications, datasets, cases of harm, and fairness evaluation resources.
2) building and testing the SVEva Fair Open Source Audit Tool, which assists voice technology developers to assess bias in their applications.
NA
¿Cómo vas a hacer frente si varía el número de participantes en tu sesión? ¿Y si asisten 30 participantes? ¿Y si son 3?:We are hosting the discussion in the ‘fishbowl format’. Our MozFest virtual space (e.g. in SpatialChat) will be designed as a Finnish sauna. Symbolically, a sauna offers intimacy and equality, however saunas have become unreachable spaces during Covid. With the virtual sauna-fishbowl we want to re-open intimacy and open dialogue.
The fishbowl works as follows. 3 panelists and a free towel are seated on a bench in the sauna. Any audience member can take the free seat on the towel and participate in the discussion alongside the panelists. After they have made their contribution they make space for another audience member to join. Audience members can stay in the sauna-fishbowl as long as they like, or listen in the background if they do not want to participate actively. If the seat on the sauna bench is empty the discussion continues as planned until someone sits down.
¿Qué pasará después del MozFest? Esperamos que muchos esfuerzos y discusiones continúen después de MozFest. Comparte cualquier idea que tengas sobre cómo continuar el trabajo de tu sesión.:The Fair EVA project is part of the MozFest Trustworthy AI Working Group, and a recipient of the Mozilla Technology Fund Award. We meet weekly to align on project progress, check in on tasks and welcome new contributors. All outputs of the MozFest session will feed directly into the Fair EVA project.
¿En qué idioma te gustaría realizar tu sesión?:English