Projects

Efficient Self-Supervised Learning for Inclusive and Innovative Speech Technologies Self-Supervised Learning (SSL) has recently emerged as an incredibly promising artificial intelligence (AI) method. Through this method, massive amounts of unlabeled data that are accessible can be utilized by AI systems to surpass known performances. Particularly, the field of Automatic Speech Processing (ASP) is swiftly being transformed by the arrival of SSL, thanks in part to massive industrial investments and the explosion of data, both provided by a handful of companies. The performance gains are impressive, but the complexity of SSL models requires researchers and industry professionals in the field to have extraordinary computational capacity, drastically limiting access to fundamental research in this area and its deployment in everyday products. For instance, a significant portion of work using an SSL model for ASP relies on a system maintained and provided by a single company (wav2vec 2.0). The entire lifecycle of the technology, from its theoretical foundations to its practical deployment and societal analysis, therefore depends solely on institutions with the physical and financial means to support the intensity of this technique’s development. The E-SSL project aims to restore to the scientific community and ASP industry the necessary control over self-supervised learning Plus d'infos

Development of a methodology for evaluating voice identification systems The BRUEL project concerns the evaluation/certification of voice identification systems against adversarial attacks. Indeed, speaker recognition systems are vulnerable not only to speech artificially produced by voice synthesis but also to other forms of attacks such as voice identity conversion and replay attacks. The artifacts created during the creation or manipulation of these fraudulent attacks leave marks in the signal by voice synthesis algorithms, thus distinguishing the original real voice from a forged voice. Under these conditions, detecting identity theft requires evaluating identity theft countermeasures concurrently with speaker recognition systems. The BRUEL project aims to propose the first methodology for evaluating/certifying voice identification systems based on a Common Criteria approach. List of partners: CEA Eurecom Service National de Police Scientifique IRCAM LIA Laboratoire d’Informatique d’Avignon Project Coordinator: LIA Scientific Manager for LIA: Driss Matrouf Start Date: 01/01/2023 End Date: 30/06/2026 More

Stream Learning for Multilingual Knowledge Transfer The internet contains vast amounts of data and information in various languages, both written and audiovisual. There’s an increasing need to leverage this largely untapped resource. The SELMA project, funded by the EU, focuses on ingesting and monitoring large quantities of data. It systematically trains machine learning models to perform tasks in natural language and utilizes these models to monitor data streams, aiming to enhance multilingual media monitoring and real-time content production. Ultimately, the project will advance cutting-edge techniques in language modeling, automatic translation, speech recognition, and synthesis. Project Coordinator: Deutsche Welle, DE Scientific Lead for LIA: Yannick ESTEVE Start Date: 01/01/2021 End Date: 30/12/2023 More

MUlti-party perceptually-active situated DIALog for human-roBOT interaction In muDialBot, our ambition is to proactively incorporate human-like behavioral traits in human-robot spoken communication. We aim to reach a new stage in harnessing the rich information provided by audio and visual data streams from humans. In particular, extracting verbal and non-verbal events should enhance the decision-making abilities of robots to manage turns of speech more naturally and also switch from group interactions to face-to-face dialogues according to the situation. There has been growing interest recently in companion robots capable of assisting individuals in their daily lives and effectively communicating with them. These robots are perceived as social entities, and their relevance to health and psychological well-being has been highlighted in studies. Patients, their families, and healthcare professionals will better appreciate the potential of these robots as certain limitations are quickly overcome, such as their ability to move, see, and listen to communicate naturally with humans, beyond what touchscreen displays and voice commands already enable. The scientific and technological outcomes of the project will be implemented on a commercial social robot and tested and validated with multiple use cases in the context of a day hospital unit. Large-scale data collection will complement in-situ Plus d'infos

Exchanges for SPEech ReseArch aNd TechnOlogies Speech processing technologies are crucial for numerous commercial applications. The ESPERANTO project, funded by the EU, aims to make the next generation of AI algorithms used in speech processing applications more accessible. For instance, they should consider human involvement and be interpretable to allow sensitive applications and safeguard personal data. ESPERANTO envisions disseminating these technologies across European SMEs, expanding and ensuring their implementation for forensic, healthcare, and educational purposes. The project will support the development of freely accessible tools, conduct seminars on various speech processing themes to assist new students, researchers, and engineers working in speech AI, and contribute to the collection and sharing of linguistic and speech-related resources. Project Coordinator: University of Le Mans, FR Scientific Manager for LIA: Jean-François Bonastre Start Date: 01/01/2021 End Date: 30/06/2025 More

AISSPER: Artificial Intelligence for Semantically controlled SPEech undeRstanding Artificial Intelligence (AI) holds strategic importance at the national level due to impressive outcomes achieved by deep learning algorithms in various domains such as natural language processing (NLP), medicine, and political analytics across a wide range of applications. France has emerged as a leader in deep learning owing to recent political efforts highlighted in recent years. Over the last decade, substantial efforts have been dedicated to end-to-end Spoken Language Understanding (SLU) systems, driven by the feasibility of applications like personal assistants and conversational systems. Superior results have been observed in automatic speech recognition (ASR) with architectures based on hyper-complex number algebra called quaternions, requiring less processing time (Morchid 2018) and fewer parameters to estimate compared to conventional models (Parcollet et al 2018; 2019). Reducing model parameters efficiently trains neural architectures with limited data quantities, often challenging to obtain for specific semantic concepts and contexts from specific domains. Intrinsically linked learning processes like ASR and SLU hinder the parallelization of learning examples, critical for lengthy sequences as memory constraints limit batch processing using examples. Furthermore, error analysis conducted on completed projects like M2CR, JOKER, VERA, SUMACC, Media, or DECODA highlighted the importance of Plus d'infos