GeezerVison is showing www.interspeech2022.org — INTERSPEECH 2022
Lin-shan Lee Taiwan University Chair Professor, National Taiwan University Title: From Semantics to Self-supervised Learning for Speech and Beyond Keynote 1 Monday 19 September 09:30 KST at Grand Ballroom Abstract Although spoken language understanding has been extensively investigated for many years, the gap between signals and semantics in speech remains to be a challenging problem. But the semantics in speech may offer a bridge towards the realization of "a spoken version of Google". On the other hand, with the fast advances in deep learning in recent years, self-supervised learning seems to be very attractive. In this latter approach, a model is first pre-trained with a big unlabeled dataset without considering any specific application, based on which many different downstream tasks can be achieved much more easier with simpler model structure and smaller labeled dataset. In this talk, the speaker will present his thoughts and experiences on these two research areas, and show how the research paths for them seem to merge in some way. BIO Lin-shan Lee has been teaching in Electrical Engineering and Computer Science at National Taiwan University since 1979. He invented, published and demonstrated the earliest but very complete set of fundamental technologies and systems for analyzing Chinese sentences (1986-91) and synthesizing (1984-89) and recognizing (1987-97) Chinese speech with unlimited text, considering the structural features of Chinese language(monosyllable per character, limited number of distinct monosyllables, tones, sentence structures, etc.) and the extremely limited resources. He then focused his work on retrieval of speech information from the Internet, proposing a whole set of approaches offering better retrieval accuracy and efficiency. This part of work applies equally to all different languages, and was described as the stepping stones towards "a spoken version of Google" when Nature selected him in 2018 as one of the 10 "Science Stars of East Asia" in a special issue on scientific research in East Asia. He has also pioneered a wide variety of other new frontiers in spoken language technologies, with examples including structured learning for ASR, code-switched speech recognition, unsupervised discovery of acoustic tokens, audio signal embedding and unsupervised ASR. So his work was not only across different languages with very different linguistic structures, not only from TTS to ASR to retrieval to understanding, but from acoustics, linguistics to semantics, from utterances to spoken documents, and from supervised to unsupervised. He served as a member of the Board (2001-2009) of International Speech Communication Association(ISCA), the Distinguished Lecturer of IEEE Signal Processing Society (SPS) (2007-2008), and the general chair of International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2009 held in Taipei. He was elected IEEE Fellow in 1993 and ISCA Fellow in 2010. He also received the Meritorious Service Award from IEEE SPS in 2011. He received Presidential Science Prize of Taiwan in 2015 and was elected as an academician of Academia Sinica in 2016, both of which are the highest honor for scientists in Taiwan.
Yejin Choi Brett Helsel Professor at the University of Washington and Senior Research Manager at AI2 Title: David V.S. Goliath: the Art of Leaderboarding in the Era of Extreme-Scale Neural Models Keynote 2 Tuesday 20 September 08:30 KST at Grand Ballroom Abstract Scale appears to be the winning recipe in today's leaderboards. And yet, extreme-scale neural models are still brittle to make errors that are often nonsensical and even counterintuitive. In this talk, I will argue for the importance of knowledge, especially commonsense knowledge, as well as inference-time algorithms, and demonstrate how smaller models developed in academia can still have an edge over larger industry-scale models, if powered with knowledge or algorithms. First, I will introduce "symbolic knowledge distillation", a new framework to distill larger neural language models into smaller commonsense models, which leads to a machine-authored KB that wins, for the first time, over a human-authored KB in all criteria: scale, accuracy, and diversity. Next, I will highlight how we can make better lemonade out of neural language models by shifting our focus to unsupervised, inference-time algorithms. I will demonstrate how unsupervised models powered with algorithms can match or even outperform supervised approaches on hard reasoning tasks such as nonmonotonic reasoning (such as counterfactual and abductive reasoning), or complex language generation tasks that require logical constraints. Finally, I will introduce a new (and experimental) conceptual framework, Delphi, toward machine norms and morality, so that the machine can learn to reason that “helping a friend” is generally a good thing to do, but “helping a friend spread fake news” is not. BIO Yejin Choi is Brett Helsel Professor at the Paul G. Allen School of Computer Science & Engineering at the University of Washington and a senior research manager at AI2 overseeing the project Mosaic. Her research investigates a wide range of problems including commonsense knowledge and reasoning, neuro-symbolic integration, multimodal representation learning, and AI for social good. She is a co-recipient of the ACL Test of Time award in 2021, the CVPR Longuet-Higgins Prize in 2021, a NeurIPS Outstanding Paper Award in 2021, the AAAI Outstanding Paper Award in 2020, the Borg Early Career Award in 2018, the inaugural Alexa Prize Challenge in 2017, IEEE AI's 10 to Watch in 2016, and the ICCV Marr Prize in 2013.
Rupal Patel Vice President, Voice & Accessibility, Veritone Title: "Blurring the line between human and computer-generated speech: opportunities and challenges" Keynote 3 Wednesday 21 September 08:30 KST at Grand Ballroom Abstract Computer-generated speech is becoming nearly indistinguishable from human recorded speech, especially for short-form audio content. The neural models that generate life-like speech require a fraction of the training data compared to previous concatenative and parametric methods. Moreover, combining text-to-speech and speech conversion methodologies can further blur the line between real and artificial. This talk will shed light on the shift in scientific focus from striving for human parity to protecting identity, safeguarding usage, and sometimes even introducing artifacts to convey provenance. Through a series of use cases, we will explore the opportunities and challenges in this new dawn of humanized speech technologies. BIO Dr. Rupal Patel is an internationally renowned speech scientist turned entrepreneur, bringing decades of clinical, academic, scientific and social entrepreneurship experience to Veritone. As Vice President of Voice & Accessibility, Dr. Patel is responsible for setting strategy and leading innovation efforts in the voice AI space, in addition to expanding the reach and impact of Veritone’s voice solutions for those living with disabilities or inequities. A preeminent thought leader in voice AI, Dr. Patel advocates for ethical transparent and fair use policies that can broaden the monetization capabilities for voice-over artists by leveraging AI. Prior to Veritone, Dr. Patel was the Founder and CEO of VocaliD, a voice AI company acquired by Veritone in 2022, that creates synthetic voices with personality for discerning brands that understand the power of customized voice and for individuals living with speechlessness who want to be heard in a voice that is uniquely theirs. VocaliD was a spinout from Dr. Patel’s research lab at Northeastern University, where she is a tenured Full Professor with interdisciplinary appointments in the Bouve College of Health Science and Khoury College of Computer Science. Named one of Fast Company’s 100 Most Creative People in Business, she has been featured on TED, NPR, WIRED, and in major international news and technology publications.
Daniel Dongyuel Lee Executive Vice President at Samsung Research and Tisch University Professor at Cornell Tech Title: “Representations and Geometry for Multimodal Learning” Keynote 4 Thursday 22 September 08:30 KST at Grand Ballroom Abstract The advent of deep neural networks has brought significant advancements in the development and deployment of speech and multimodal technologies. Compared with conventional signal processing approaches, end-to-end neural network architectures have shown significantly better performance for natural language understanding, machine translation, automatic speech recognition, and text-to-speech generation. Samsung Research has focused on incorporating these neural network models across Samsung's billions of devices and users. But how can we understand how the representations of sensor input signals are transformed by deep neural networks? I show how insights can be gained by analyzing the high-dimensional geometrical structure of these representations as they are reformatted in neural network hierarchies. BIO Dr. Daniel Dongyuel Lee is the Tisch University Professor in Electrical and Computer Engineering at Cornell Tech and Executive Vice President and Head of the Global AI Center for Samsung Research. He received his B.A. summa cum laude in Physics from Harvard University and his Ph.D. in Condensed Matter Physics from the Massachusetts Institute of Technology. He was also a researcher at Bell Labs in the Theoretical Physics and Biological Computation departments. He is a Fellow of the IEEE and AAAI and has received the NSF CAREER award and the Lindback award for distinguished teaching. He was also a fellow of the Hebrew University Institute of Advanced Studies in Jerusalem, an affiliate of the Korea Advanced Institute of Science and Technology, and organized the US-Japan National Academy of Engineering Frontiers of Engineering symposium and Neural Information Processing Systems (NeurIPS) conference. His group focuses on understanding general computational principles in biological systems and on applying that knowledge to build autonomous systems.
Top
|
