Keynote 1 From Minsky's Society of Mind to a Society of Minds? Artificial Intelligence, Language Models and Agency with Joscha Bach

Abstract: Artificial Intelligence is mostly concerned with creating intelligent systems by imposing order on mindless parts. Marvin Minsky's seminal book "Society of Mind" suggests that a mind is composed of a multitude of interacting agents, coordinating to create the overall functionality of intelligent agency. If we extend this paradigm beyond individual minds, we are entering the realm of collective, social agency, and can discuss about ethics based on shared purposes.

Privacy and security using speech and speaker recognition techniques

• Federated Learning for Privacy Preserving On-Device Speaker Recognition
  Abraham Woubie and Tom Bäckström
• The Use of Audio Fingerprints for Authentication of Speakers on Speech Operated Interfaces
  Abraham Woubie, Tom Bäckström and Pablo Pérez Zarazaga
• Graph2Speak: Improving Speaker Identification using Network Knowledge in Criminal Conversational Data
  Maël Fabien, Seyyed Saeed Sarfjoo, Petr Motlicek, Srikanth Madikeri
• Visualizing Automatic Speech Recognition – Means for a Better Understanding?
  Karla Markert, Romain Parracone, Mykhailo Kulakov, Philip Sperl, Ching-Yu Kao, Konstantin Böttinger

Speech privacy, speaker anonymization and legal regulations

• Evaluation of Speaker Anonymization on Emotional Speech
  Hubert Nourtel, Pierre Champion, Denis Jouvet, Anthony Larcher, Marie Tahon
• Brave New World? Processing of personal data about employees under Art. 9 of GDPR in the context of human-robot interaction
  Ivo Emanuilov, Katerina Yordanova
• A More Holistic View of Literacy Specifications in the Context of Speech-based Systems
  Carolin Wienrich, Astrid Carolus
• Design Implications for Human-Machine Interactions from a Qualitative Pilot Study on Privacy
  Anna Leschanowsky,Birgit Brüggemeier, Nils Peters

Keynote 2 Racial Disparities in Automated Speech Recognition with Allison Koenecke

Abstract: Automated speech recognition (ASR) systems are now used in a variety of applications to convert spoken language to text, from virtual assistants, to closed captioning, to hands-free computing. By analyzing a large corpus of sociolinguistic interviews with white and African American speakers, we demonstrate large racial disparities in the performance of popular commercial ASR systems developed by Amazon, Apple, Google, IBM, and Microsoft. Our results point to hurdles faced by African Americans in using increasingly widespread tools driven by speech recognition technology. More generally, our work illustrates the need to audit emerging machine-learning systems to ensure they are broadly inclusive. See more at fairspeech.stanford.edu.

Privacy and security using speech and speaker recognition techniques

• Open-Set Speaker Identification pipeline in live criminal investigations
  Maël Fabien, Petr Motlicek
• Language Dependencies in Adversarial Attacks on Speech Recognition Systems
  Karla Markert, Donika Mirdita, Konstantin Böttinger
• ROXSD: a Simulated Dataset of Communication in Organized Crime
  Kvetoslav Maly, Gerhard Backfried, Francesco Calderoni, Jan "Honza" Černocký, Erinc¸ Dikici, Maël Fabien, Jan Hořínek, Joshua Hughes, Miroslav Janošík, Marek Kovac, Petr Motlicek, Hoang H. Nguyen, Shantipriya Parida, Johan Rohdin, Miroslav Skácel, Sergej Zerr, Dietrich Klakow, Dawei Zhu, Aravind Krishnan
• Robustifying automatic speech recognition by extracting slowly varying features
  Matias Pizarro, Dorothea Kolossa, Asja Fischer

Speech privacy, speaker anonymization and legal regulations

• Revisiting Speech Content Privacy
  Jennifer Williams, Junichi Yamagishi, Paul-Gauthier, Cassia Valentini-Botinhao, Jean-François Bonastre
• Implications of the New Regulation Proposed by the European Commission on Automatic Content Moderation
  Vera Schmitt, Veronika Solopova, Vinicius Woloszyn, Jessica de Jesus de Pinho Pinhal
• Benchmarking and challenges in security and privacy for voice biometrics
  Jean-François Bonastre, Héctor Delgado, Nicholas Evans, Tomi Kinnunen, Kong Aik Lee, Xuechen Liu, Andreas Nautsch, ‪Paul-Gauthier Noé‬, Jose Patino, Md Sahidullah, Brij Mohan Lal Srivastava, Massimiliano Todisco, Natalia Tomashenko, Emmanuel Vincent, Xin Wang, Junichi Yamagishi
• Intuitive Privacy from Acoustic Reach: A Case for Networked Voice User-Interfaces
  Tom Bäckström, Sneha Das, Pablo Pérez Zarazaga, Johannes Fischer, Rainhard Dieter Findling, Stephan Sigg and Le Ngu Nguyen
• Speaker anonymization solution for public voice-assistant interactions – Presentation of a Work in Progress Development
  Ingo Siegert

Keynote 3 Exploiting the Gaps Between Human and Machine Understanding of Audio: Frameworks, Attacks, and Defenses with Patrick Traynor

Abstract: Modern machine learning techniques now enable a wide range of voice-driven systems. Such systems not only power our personal assistants and transcribe our text message, but also enable the creation of convincing virtual avatars, assist in air traffic control, and give voice to those who can no longer speak. However, the algorithms underlying these systems process and "understand" audio far differently that humans do, creating substantial vulnerabilities. In this talk, I discuss a range of such attacks and how they target real systems, a shared framework by which these attacks can be compared, and how such vulnerabilities might actually serve as the basis of stronger systems.