Keynotes

 

Hervé Glotin

Université de Toulon, CNRS, LIS

Stereophonic autoencoder on bio-transient to enhance biodiversity tracking, and other methods to monitor abyssea

 

Abstract

Soundscape analysis is a powerful method to assess biodiversity. We propose a new paradigm of stereophonic deep learning (bio)acoustic representation, taking advantages of stereophonic soundscape big data. This new model allows to elaborate efficient representation of (bio)transient jointly to enhance estimation of their Time Delay of Arrival. We present simulations and real data analysis on Physeter macrocephalus. We finally demonstrate the first 3D tracking of this animal using passive acoustics from a short hydrophone array placed on an Autonomous Surface Vehicle. We discuss on details of the use of its biosonar and give perspectives on abyssal biodiversity survey.

 

Biography

Hervé Glotin is professor of Computer Science at LIS CNRS & Univ. of Toulon (UTLN) since 2010. He is honorary member of the Institut Universitaire de France (IUF). He is leading the DYNI team on stochastic multimodal information retrieval. He received in 2001 his PhD 'Robust adaptive multi-stream automatic speech recognition using voicing and localization cues' from Inst. of Perceptual Artificial Intelligence (IDIAP-CH) and Inst. of Spoken Communication Grenoble (INPG-FR). He got his master in artificial intelligence from Paris Sorbonne and INPG, during which he initiated a model of vocalic system evolution and emergence of a phonetic code in a society of communicating agents. In 2000 he was expert at Johns Hopkins CSLP Lab. with IBM human language team to co-design the Via-Voice audiovisual Large Vocabulary Speech Recognizer. From 2001 to 2003 he was CNRS Research Engineer in stochastic semantics (Toulouse) and became Associate Professor at UTLN in 2003. His research deals with machine learning and information retrieval including scaled acoustic biodiversity. He is since 10 years the general chair of ERMITES workshop on multimodal information retrieval. He was the general chair of the workshops ICML2013, NIPS2013, ICML2014, ICDM2015 on machine learning for bioacoustics, where he organised the first bird song classification challenges (100 species). He was the local chair of ICLR 2017 (1500 participants). He is currently co-pi of the Bird LifeCLEF challenges (1500 species since 2016). He initiated in 2012 and heads the CNRS Big Data project on Scaled Acoustic Biodiversity (http://sabiod.org), involving several expert teams in machine learning, signal processing and bioacoustics. He is co-author of more than 100 of internat. refereed articles, and of patents on real-time bioacoustic indexing. He regularly serves on numerous scientific review in international journals (PLOS One, IEEE Trans. on Multimedia, MTAP, IEEE Trans Sensors...) and conferences.

 

Hanna Lukashevich

Fraunhofer Institute for Digital Media Technology IDMT

Acoustic condition monitoring for smart city and industry environments

 

Abstract

Sound is surrounding us; it is a natural source of information. Humans and animals use sound for communication, it helps to recognize danger even while sleeping. Also the urban and industrial environments are full of intended and unintended sounds. With the state-of-the-art intelligent acoustic measurement systems we can make use of these sounds and gain information that help us to make cities smarter and production sites more efficient. The talk will cover multiple use cases of acoustic condition monitoring in smart city and industry environments. In addition to the algorithmic approach of information retrieval, based on digital signal processing and machine learning, the talk will address the implementation aspects, covering the requirements analysis with customer constraints; privacy and security; challenges on data acquisition and data annotation; model quality and acceptance; deployment and integration.

 

Biography

Hanna Lukashevich is head of the Semantic Music Technologies (SMT) research group and Industrial Media Applications (IMA) business unit at Fraunhofer Institute for Digital Media Technology IDMT in Ilmenau, Germany. Among her research interests are audio signal processing and machine learning. As head of the SMT group and IMA business unit, she manages a number of R&D projects, both for companies in the industrial and digital media sector as well as national and international public projects.