National audienceSpeaker personalized acoustic models are obtained from a global model by updating its parameters using speaker's data. An important question is whether access to these personalized models allows to easily build an attack to identify the associated speaker. This problem is especially important in the context of federated learning of speech recognition acoustic models where a global model is learnt on the server using the updates received from multiple clients. We propose an approach to analyze information in neural network acoustic models based on a neural network footprint on the so-called indicator dataset. Using this method, we develop two very effective attack models that allow to infer speaker identity from the updat...
This study proposes a security/surveillance system capable of automatically recognizing and detectin...
Deepfakes are synthetically generated media often devised with malicious intent. They have become in...
Speaker verification is the process of authenticating a person’s identity. Most of the available spe...
National audienceSpeaker personalized acoustic models are obtained from a global model by updating...
International audienceThis paper investigates methods to effectively retrieve speaker information ...
International audienceThe widespread of powerful personal devices capable of collecting voice of the...
International audienceThe widespread of powerful personal devices capable of collecting voice of the...
International audienceThis paper investigates different approaches in order to improve the performan...
Voice biometric systems are being used in various applications for secure user authentication using ...
International audienceIn recent years, deep learning breakthroughs met a huge success in automatic s...
Model extraction is the process of reverse-engineering the parameters of a machine learning model ex...
The speech signal conveys information about the identity of the speaker. The area of speaker identif...
The performance of speaker recognition systems has considerably improved in the last decade. This is...
Texte intégral accessible uniquement aux membres de l'Université de LorraineIn this thesis we focus ...
Biometric security systems based on predefined speech sentences are extremely common nowadays, parti...
This study proposes a security/surveillance system capable of automatically recognizing and detectin...
Deepfakes are synthetically generated media often devised with malicious intent. They have become in...
Speaker verification is the process of authenticating a person’s identity. Most of the available spe...
National audienceSpeaker personalized acoustic models are obtained from a global model by updating...
International audienceThis paper investigates methods to effectively retrieve speaker information ...
International audienceThe widespread of powerful personal devices capable of collecting voice of the...
International audienceThe widespread of powerful personal devices capable of collecting voice of the...
International audienceThis paper investigates different approaches in order to improve the performan...
Voice biometric systems are being used in various applications for secure user authentication using ...
International audienceIn recent years, deep learning breakthroughs met a huge success in automatic s...
Model extraction is the process of reverse-engineering the parameters of a machine learning model ex...
The speech signal conveys information about the identity of the speaker. The area of speaker identif...
The performance of speaker recognition systems has considerably improved in the last decade. This is...
Texte intégral accessible uniquement aux membres de l'Université de LorraineIn this thesis we focus ...
Biometric security systems based on predefined speech sentences are extremely common nowadays, parti...
This study proposes a security/surveillance system capable of automatically recognizing and detectin...
Deepfakes are synthetically generated media often devised with malicious intent. They have become in...
Speaker verification is the process of authenticating a person’s identity. Most of the available spe...