Laboratoire Informatique d'Avignon

What Laboratoire Informatique d'Avignon does

Laboratoire Informatique d'Avignon (LIA, UPR 4128) is the computer science research laboratory of Avignon Université, France, founded in 1987 by Henri Méloni and currently directed by Fabrice Lefèvre. The lab brings together approximately 90 members — 29 permanent faculty and staff, roughly 40 doctoral students, plus postdoctoral researchers, engineers, and master's interns — organized into two teams: SLG (Speech and Language Group) and CORNET (computer networks and operational research). Its mission is fundamental and applied research in automatic speech and language processing, computer networks, and operational research, serving the academic research community and, through the LIAvignon industrial AI chair, selected industry and government partners.

LIA's core technical outputs are open-source models, toolkits, and datasets rather than commercial products. The flagship assets include SpeechBrain, a PyTorch-based all-in-one speech toolkit launched in 2020 in collaboration with MILA and sponsored by Samsung, Dolby, and Nvidia; DrBERT, a French biomedical language model based on RoBERTa and pre-trained on the proprietary NACHOS corpus, evaluated across 11 biomedical NLP tasks; and novel architectures such as quaternion neural networks for multidimensional signal data and FedStaleAsync for asynchronous federated learning. The lab also produces specialized research datasets (Serial Speakers, WAC) and operates in adjacent areas including anti-spoofing, speaker verification quantization, speech-to-speech translation (Hibiki), biomedical NLP, and network resource optimization. AI modalities covered span text, audio, multimodal, and structured data.

The business model is non-commercial and entirely publicly funded. Revenue mechanics are ANR (Agence Nationale de Recherche) competitive grants, institutional support from Avignon Université, membership-based funding from research federations (Carnot Cognition Institute, BLRI Labex, Agorantic FR 3621, Tersys SFR 4241), and industry collaboration through the LIAvignon AI Chair. Concurrent active projects include BRUEL (2023-2026, voice identification evaluation with police scientifique), ESSL (efficient self-supervised learning), PANTAGRUEL (multimodal French LLMs), and MALADES (LLM deployment in health centers). The lab has no pricing model, no commercial customers, and does not disclose revenue; distribution of research outputs occurs through GitHub, peer-reviewed publication, and conference presentations rather than commercial channels.

Name

Laboratoire Informatique d'Avignon

Legal name

Laboratoire Informatique d'Avignon

Website

https://lia.univ-avignon.fr

Company type

Private

Founded year

1987

Operating status

Operating

Headcount range

501–1,000 employees

Short description

Laboratoire Informatique d'Avignon (LIA) is a French university research laboratory (UPR 4128) at Avignon Université conducting foundational research in automatic speech and language processing, computer networks, and operational research, producing open-source tools like SpeechBrain and DrBERT for the global research community.

Ownership category

akta.pro rank

Product category

Academic Research in Speech and Language Processing

NAICS

Scientific Research and Development Services (5417)

SIC

Services-Educational Services (8200)

akta.pro primary industry

NLP Developer Tools & Evaluation (prompting, testing, benchmarks) (HDAAADAK)

Keywords

• Automatic speech processing
• Natural language processing
• Speech recognition research
• Biomedical language models
• Federated learning research

Headquarters

HQ city

Avignon

HQ country

France

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Services

Cost components

Personnel, Technology or R&D, Operations

Revenue model

1. Public Research Funding: Funded primarily through ANR (Agence Nationale de Recherche) grants, institutional support from Avignon University, and membership in research federations such as Carnot Cognition Institute, Brain and Language Research Institute (BLRI), Agorantic, and Tersys
2. Industrial Partnership - LIAvignon AI Chair: Hosts the LIAvignon industrial partnership chair in Artificial Intelligence with focus on voice, speech, and speaker processing, providing collaborative research with industry partners

Go-to-market motion1 record

Distribution channels3 records

Marketing channels4 records

Core offering

Laboratoire Informatique d'Avignon (LIA) is a public university research laboratory (UPR 4128) that conducts fundamental and applied research in automatic speech processing, natural language processing, computer networks, and operational research. Its primary outputs include peer-reviewed publications, open-source software toolkits (SpeechBrain), pre-trained AI models (DrBERT), and annotated research datasets (NACHOS, Serial Speakers, WAC), distributed through academic channels and GitHub.

Product overview

LIA (Laboratoire Informatique d'Avignon) is a university research laboratory (UPR 4128) that does not operate commercial products but produces research outputs including open-source tools, datasets, and pretrained models. The main offerings include DrBERT (French biomedical language model), SpeechBrain (open-source PyTorch-based speech toolkit), NACHOS biomedical corpus, and various research datasets (Serial Speakers, WAC). The laboratory is organized into two research teams: CORNET (networks, optimization, games) and SLG (speech and language processing), with a focus on speech/audio processing, natural language processing, computer networks, and operational research. Research outputs are made available as open-source software on GitHub and datasets for academic collaboration.

Differentiator

Problem solved

Functional benefit

Brands

• CORNET: Research team focused on computer networks and operational research at LIA.
• SLG (Speech and Language Group)
• LIAvignon AI Chair

Products and services

• SpeechBrain Open-source PyTorch-based all-in-one speech toolkit for developing automatic speech recognition (end-to-end and DNN-HMM), speaker recognition, speech separation, multi-microphone processing, and self-supervised learning systems, developed in collaboration with MILA.
• DrBERT French biomedical language model based on RoBERTa architecture, pre-trained on the NACHOS French biomedical corpus, evaluated across 11 French biomedical NLP tasks including named entity recognition, POS tagging, classification, and question answering.
• NACHOS Corpus French biomedical text corpus used to pre-train the DrBERT language model, available exclusively for academic research purposes.
• Serial Speakers Dataset Collection of manually annotated TV series including Breaking Bad (S01-S05), Game of Thrones (S01-S08), and House of Cards (S01-S02) for speaker recognition and audiovisual research.
• WAC (Wikipedia Abusive Conversations) Dataset containing Wikipedia editor conversations annotated for various types of abuse at the message level, aligning WikiConv messages with Wikipedia Comment Corpus toxicity annotations.
• LIAvignon AI Chair Industrial partnership chair in Artificial Intelligence hosted by LIA, focused on voice, speech, and speaker processing; serves as the laboratory's flagship vehicle for applied collaboration with industrial partners.

Quantifiable outcome

• Macro EER reduced from 5.46% to 4.81% (11.9% relative improvement) for anti-spoofing using Mixture-of-Experts architecture
• +2 more outcomes

Named customers3 records

Segments3 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability13 records

Feature6 records

Partnerships

14 partnerships are on record, tiered core, flagship, major and minor.

• Avignon UniversitécoreStrategic or Co-development Partner · 1 January 1987Primary institutional affiliation. LIA is the Computer Science Laboratory of Avignon University (UPR 4128), integrated into the CERI department. The university provides faculty members, administrative support, and infrastructure for the laboratory's research activities.
• Carnot Cognition InstitutecoreStrategic or Co-development PartnerLIA is a member of the Carnot Cognition Institute, a network of French research laboratories focused on cognitive sciences and artificial intelligence, providing access to collaborative research opportunities and institutional recognition.
• Brain and Language Research Institute (BLRI)coreStrategic or Co-development PartnerMember of the Brain and Language Research Institute (BLRI) Laboratory of Excellence (Labex), participating in research federation activities related to language and cognitive research.
• Agorantic Research Federation (FR 3621)coreStrategic or Co-development PartnerParticipates in the Agorantic research federation (FR 3621), collaborating with other research institutions on interdisciplinary projects and hosting doctoral contracts through this federation.
• Tersys Research Structure (SFR 4241)coreStrategic or Co-development PartnerParticipates in the Tersys federative research structure (SFR 4241), contributing to collaborative research initiatives with partner institutions.
• LIAvignon (AI Chair)flagshipStrategic or Co-development PartnerIndustrial partnership chair in Artificial Intelligence hosted by LIA, with particular focus on voice, speech, and speaker processing. Represents the laboratory's flagship industry collaboration vehicle.
• CEAmajorStrategic or Co-development PartnerPartner in the ANR BRUEL project for developing evaluation and certification methodology for voice identification systems. CEA brings expertise in technology and security assessment.
• EurecommajorStrategic or Co-development PartnerPartner in the ANR BRUEL project, contributing research expertise in signal processing and security for voice identification system evaluation.
• IRCAMmajorStrategic or Co-development PartnerPartner in the ANR BRUEL project, bringing expertise in sound analysis and acoustic modeling for voice identification evaluation.
• Service National de Police ScientifiquemajorStrategic or Co-development PartnerPartner in ANR BRUEL project, providing practical requirements for voice identification systems in forensic science and police investigations.
• Université Paris Dauphine (LAMSADE)minorStrategic or Co-development PartnerPartner in the ANR ESSL project for efficient self-supervised learning in speech technologies, contributing to computational efficiency research.
• Laboratoire d'Informatique de GrenobleminorStrategic or Co-development PartnerPartner in the ANR ESSL project, collaborating on self-supervised learning research for speech technologies.
• MILA (Montreal Institute for Learning Algorithms)minorStrategic or Co-development PartnerCollaborated in the SpeechBrain project launch, with Dr. Mirco Ravanelli from MILA leading the initiative. Represents international AI research collaboration.
• Samsung, Dolby, NvidiaminorStrategic or Co-development PartnerInitial industry sponsors supporting the SpeechBrain open-source speech toolkit development, providing computing resources and industrial perspective.

Scale indicators4 records

Recent moves6 records

Expansion highlights5 records

Direct peers

• INRIA: France's national computer science research institute operating research teams in speech, NLP, and AI. Directly comparable as a publicly funded French CS research organization with similar ANR-funded project structure and publication output.
• IRCAM: Paris-based institute for music and acoustics research, existing BRUEL project partner. Comparable French public research entity with overlapping expertise in audio/speech analysis and acoustic modeling.
• LIG (Laboratoire d'Informatique de Grenoble): Grenoble computer science research lab and existing ANR ESSL partner. Comparable multi-team French CS research lab with overlapping NLP and machine learning focus areas.
• LIFAT (Tours): University of Tours computer science lab working on NLP, machine learning, and computer vision. Comparable French university research lab with shared ANR project culture.
• LIUM (Le Mans University): University of Le Mans computer science laboratory with closely aligned focus on automatic speech recognition, speaker recognition, NLP, and machine translation. Highly comparable in mission, scale, and funding model.
• GIPSA-Lab (Grenoble): Multidisciplinary French research lab focused on speech, signal processing, and control. Highly comparable in speech-processing research depth and CNRS/Université funding structure.
• IRIT (Toulouse): Toulouse Institute of Computer Science Research, a large French university CS lab with speech, NLP, and AI teams. Comparable scale (~90+ members), funding mix (ANR), and academic mission.
• LISN (formerly LIMSI, CNRS): CNRS-affiliated computer science lab with one of France's strongest speech processing groups (speaker recognition, ASR). Direct peer in core SLG team mission and historical positioning.

Broad incumbents

• Naver Labs Europe: Industrial AI research lab with European base working on speech, multimodal AI, and NLP. Comparable research areas but operating with commercial funding and productization mandate.

Emerging players

• MILA (Quebec AI Institute): Montreal-based AI research institute that co-developed SpeechBrain with LIA. Comparable as a leading speech/AI research entity, though larger, English-anchored, and more internationally funded.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat5 records

Key risks5 records

Key highlights6 records

Customer concentration

• Xtwitter.com/LIAvignon
• LinkedInlinkedin.com/company/universit-davignon-lia

Revenue estimate

Valuation estimate

Number of profiles

Profiles1 record

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Laboratoire Informatique d'Avignon