Cost-effective, Multilingual, Privacy-driven voice-enabled Services

Deliverables

Open Research Data Pilot (1)

Initial data management plan

Initial data management plan

Documents, reports (4)

Initial scientific evaluation

First combined evaluation of baseline speech, dialog, and translation tools.

Platform hardware and software architecture

Platform specification including main requirements, hardware and software architecture

SDK software architecture

SDK specification including main requirements and software architecture

Data protection and GDPR requirements

Guidelines, procedures and recommendations how to implement personal data protection in the platform.

Websites, patent fillings, videos etc. (2)

Dissemination and communication action plan

Summarises all planned dissemination actions and provides communication material to the partners (logo, graphical chart, public website, short presentation of the project, templates, poster, leaflet)

First dissemination and communication report

Update on the actions conducted and web document defining the COMPRISE knowledge repository on multilingual voice-enabled applications.

Other (5)

Baseline speech and text transformation and model learning library

Design, implementation, and evaluation of baseline transformations focusing on deleting the user’s identity and words carrying critical information, and model learning.

Improved transformation library and initial privacy guarantees

Design, implementation, and evaluation of speech and text transformations addressing more types of private information and initial statistical utility/privacy bounds.

Initial multilingual interaction library

Software components and documentation for speech-to-speech translation and integration of dialog systems in the operating branch.

Initial weakly supervised learning library

Design, implementation, and evaluation of weakly supervised learning for spoken language understanding.

Initial personalised learning library for speech-to-text

Design, implementation, and evaluation of initial model personalisation strategies for speech-to-text.

Publications

Conference proceedings (15)

Privacy-Preserving Adversarial Representation Learning in ASR: Reality or Illusion?

Author(s): Srivastava, Brij Mohan Lal; Bellet, Aurélien; Tommasi, Marc; Vincent, Emmanuel
Published in: INTERSPEECH 2019, Issue 15/09/2019, 2019, Page(s) 3700-3704

Introducing the VoicePrivacy initiative

Author(s): Tomashenko, Natalia; Srivastava, Brij Mohan Lal,; Wang, Xin; Vincent, Emmanuel; Nautsch, Andreas; Yamagishi, Junichi; Evans, Nicholas; Patino, Jose; Bonastre, J.-F; Noé, Paul-Gauthier; Todisco, Massimiliano
Published in: INTERSPEECH 2020, Issue 25/10/2020, 2020

Evaluating Voice Conversion-based Privacy Protection against Informed Attackers

Author(s): Srivastava, Brij Mohan Lal; Vauquier, Nathalie; Sahidullah, Md; Bellet, Aurélien; Tommasi, Marc; Vincent, Emmanuel
Published in: 2020 IEEE International Conference on Acoustics, Speech, and Signal Processing, Issue 04/05/2020, 2020, Page(s) 2802-2806

Investigating the Impact of Pre-trained Word Embeddings on Memorization in Neural Networks

Author(s): Thomas, Aleena; Adelani, David; Davody, Ali; Mogadala, Aditya; Klakow, Dietrich
Published in: 23rd International Conference on Text, Speech and Dialogue, Issue 08/09/2020, 2020

Design Choices for X-vector Based Speaker Anonymization

Author(s): Srivastava, Brij Mohan Lal; Tomashenko, Natalia; Wang, Xin; Vincent, Emmanuel; Yamagishi, Junichi; Maouche, Mohamed; Bellet, Aurélien; Tommasi, Marc
Published in: INTERSPEECH 2020, Issue 25/10/2020, 2020

The COMPRISE Cloud Platform

Author(s): Skadiņš, Raivis; Salimbajevs, Askars
Published in: 1st International Workshop on Language Technology Platforms, Issue 16/05/2020, 2020, Page(s) 108–111

Using Privacy-Transformed Speech in the Automatic Speech Recognition Acoustic Model Training

Author(s): Salimbajevs, Askars
Published in: 9th International Conference on Human Language Technologies – the Baltic Perspective, Issue 22/09/2020, 2020

Assessing Unintended Memorization in Neural Discriminative Sequence Models

Author(s): Helali, Mossad; Kleinbauer, Thomas; Klakow, Dietrich
Published in: 23rd International Conference on Text, Speech and Dialogue, Issue 08/09/2020, 2020

Private Protocols for U-Statistics in the Local Model and Beyond

Author(s): Bell, James; Bellet, Aurélien; Gascón, Adrià; Kulkarni, Tejas
Published in: International Conference on Artificial Intelligence and Statistics, Issue 27/08/2020, 2020, Page(s) 1573-1583

Data Augmentation for Pipeline-Based Speech Translation

Author(s): Alves, Diego; Salimbajevs, Askars; Pinnis, Mārcis
Published in: 9th International Conference on Human Language Technologies – the Baltic Perspective, Issue 22/09/2020, 2020

A Comparative Study of Speech Anonymization Metrics

Author(s): Maouche, Mohamed; Srivastava, Brij Mohan Lal; Vauquier, Nathalie; Bellet, Aurélien; Tommasi, Marc; Vincent, Emmanuel
Published in: INTERSPEECH 2020, Issue 25/10/2020, 2020

On Semi-Supervised LF-MMI Training of Acoustic Models with Limited Data

Author(s): Sheikh, Imran; Vincent, Emmanuel; Illina, Irina
Published in: INTERSPEECH 2020, Issue 25/10/2020, 2020

Achieving Multi-Accent ASR via Unsupervised Acoustic Model Adaptation

Author(s): Turan, M. A. Tuğtekin; Vincent, Emmanuel; Jouvet, Denis
Published in: INTERSPEECH 2020, Issue 25/10/2020, 2020

Privacy Guarantees for De-identifying Text Transformations

Author(s): Adelani, David Ifeoluwa; Davody, Ali; Kleinbauer, Thomas; Klakow, Dietrich
Published in: INTERSPEECH 2020, Issue 25/10/2020, 2020

Who Started this Rumor? Quantifying the Natural Differential Privacy Guarantees of Gossip Protocols

Author(s): Bellet, Aurélien; Guerraoui, Rachid; Hendrikx, Hadrien
Published in: 34th International Symposium on Distributed Computing, Issue 12/10/2020, 2020

Peer reviewed articles (1)

How can Private Information Recorded by Voice-Enabled Systems be Identified?

Author(s): Moretón Poch, Álvaro; Jaramillo, Ariadna
Published in: European Data Protection Law Review, Issue to appear, 2020, ISSN 2364-2831