Abstract: Speech is a complex process emitting a wide range of biosignals, including, but not limited to, acoustics. These biosignals – stemming from the articulators, the articulator muscle activities, the neural pathways, and the brain itself – can be used to circumvent limitations of conventional speech processing in particular, and to gain insights into the process of speech production in general. In my talk I will present ongoing research at the Cognitive Systems Lab (CSL), where we explore a variety of speech-related muscle and brain activities based on machine learning methods with the goal of creating biosignal-based speech processing devices for communication applications in everyday situations and for speech rehabilitation, as well as gaining a deeper understanding of spoken communication. Several applications will be described such as Silent Speech Interfaces that rely on articulatory muscle movement captured by electromyography to recognize and synthesize silently produced speech, Brain-to-text interfaces that recognize continuously spoken speech from brain activity captured by electrocorticography to transform it into text, and Brain-to-Speech interfaces that directly synthesize audible speech from brain signals.

Speaker Bio: Tanja Schultz received her diploma and doctoral degree in Informatics from University of Karlsruhe, Germany, in 1995 and 2000. Prior to these degrees she completed the state exam in Mathematics, Sports, Physical and Educational Science from Heidelberg University, Germany in 1989. She is currently the Professor for Cognitive Systems at the University of Bremen, Germany and adjunct Research Professor at the Language Technologies Institute of Carnegie Mellon, PA USA. Since 2007, she directs the Cognitive Systems Lab, where her research activities include multilingual speech recognition and the processing, recognition, and interpretation of biosignals for human-centered technologies and applications. Prior to joining University of Bremen, she was a Research Scientist at Carnegie Mellon (2000-2007) and a Full Professor at Karlsruhe Institute of Technology in Germany (2007-2015). Dr. Schultz is an Associate Editor of ACM Transactions on Asian Language Information Processing (since 2010), serves on the Editorial Board of Speech Communication (since 2004), and was Associate Editor of IEEE Transactions on Speech and Audio Processing (2002-2004). She was President (2014-2015) and elected Board Member (2006-2013) of ISCA, and a General Co-Chair of Interspeech 2006. She was elevated to Fellow of ISCA (2016) and to member of the European Academy of Sciences and Arts (2017). Dr. Schultz was the recipient of the Otto Haxel Award in 2013, the Alcatel Lucent Award for Technical Communication in 2012, the PLUX Wireless Biosignals Award in 2011, and the Allen Newell Medal for Research Excellence in 2002, and received the ISCA / EURASIP Speech Communication Best paper awards in 2001 and 2015.

Rob Clark
Google, London, UK
Synthesizing variation in prosody for Text-to-Speech

Abstract: This talk addresses the issue of producing appropriate and engaging text-to-speech.  The quality of speech produced by modern text-to-speech systems is sufficiently intelligible and naturally sounding that we are now seeing it widely used in an increasing number of real world applications. While the speech generated can sound very natural, we are still a long way from ensuring it always sounds appropriate and engaging in the context of a particular discourse or dialogue. We present recent work at Google which begins to address this issue by looking at techniques to generate variation in prosody and speaking style using latent representations and discuss the problems and challenges that we face in going further.

Speaker Bio: Rob Clark received his PhD from the University of Edinburgh in 2003. His primary interest is in producing engaging synthetic speech. Before joining Google Rob was at the University of Edinburgh for many years involved in both teaching and research relating to text-to-speech synthesis. Rob was one of the primary developers and maintainers of the open source Festival text-to-speech synthesis system. In 2015 he joined Google where he is working on text-to-speech synthesis and prosody.

Abstract: Automatic Question Answering (Q&A), i.e., the task of building computer programs that are able to answer question posed in natural language, has a long tradition in the fields of Natural Language Processing and Information Retrieval. In recent years, Q&A applications have had a tremendous impact in industry and they are ubiquitous (e.g., embedded in any of the personal assistants that are in the market, Siri, Alexa, Cortana, Google Assistant, etc.). At the same time, we have witnessed a renewed interest in the scientific community, as Q&A has become one of the paradigmatic tasks for assessing the ability of machines to comprehend text. A plethora of corpora, resources and systems have blossomed and flooded the community in the last three years. These systems can do very impressive things, for instance, finding answers to open ended questions in long text contexts with super-human accuracy, or answering complex questions about images, by mixing the two modalities. As in many other fields, these state-of-the-art systems are implemented using machine learning in the form of neural networks (deep learning). The new AI, of course. But do these Q&A systems really understand what they read? In more simple words, do they provide the right answers for the right reasons? Several recent studies have shown that QA systems are actually very brittle. They generalize badly and they fail miserably when presented with simple adversarial examples. The machine learning algorithms are very good at picking all the biases and artefacts in the corpora, and they learn to find answers based on shallow text properties and pattern matching. But they do not show many understanding or reasoning abilities, after all. Following this serious setback, there is a new push in the community for carefully designing more complex and bias-free datasets, and more robust and explainable systems. Hopefully, this will lead to a new generation of smarter and more useful Q&A engines in the near future. In this talk, I will overview the present and the future of Question Answering by going over all the aforementioned topics.

Speaker Bio: Lluís Màrquez is a Principal Applied Scientist at Amazon Research in
Barcelona. From 2013 to 2017 he had a Principal Scientist role at the Arabic Language Technologies group from the Qatar Computing Research Institute (QCRI), and previously, he was Associate Professor at the Technical University of Catalonia (UPC, 2000-2013). He holds a university award-winning PhD in Computer Science from UPC (1999). His research focuses on natural language understanding by using statistical  machine learning models. He has 150+ papers in Natural Language Processing and Machine Learning journals and conferences. He has been General and Program Co-chair of major conferences in the area (ACL, EMNLP, EACL, CoNLL, *SEM, EAMT, etc.), and held several organizational roles in ACL and EMNLP too. He was co-organizer of various international evaluation tasks at Senseval/SemEval (2004, 2007, 2010, 2015-2017) and CoNLL shared tasks (2004-2005, 2008-2009). He was Secretary and President of the ACL special interest group on Natural Language Learning
(SIGNLL) in the period 2007-2011. More recently, he was President-elect and President of the European Chapter of the ACL (EACL; 2013-2016) and member of the ACL Executive Committee (2015-2016). Luís Màrquez has been Guest Editor of special issues at Computational Linguistics, LRE, JNLE, and JAIR in the period (2007-2015). He has participated in 16 national and EU research projects, and 2 projects with technology transfer to the industry, acting as the principal site researcher in 10 of them, helping companies embed AI in their business.