InterSpeech 2014

Achievements and Challenges of Deep Learning - From Speech Analysis And Recognition To Language And Multimodal Processing

Li Deng, Microsoft Research, Redmond, USA
Artificial neural networks have been around for over half a century and their applications to speech processing have been almost as long, yet it was not until year 2010 that their real impact had been made by a deep form of such networks, built upon part of the earlier work on (shallow) neural nets and (deep) graphical models developed by both speech and machine learning communities. This keynote will first reflect on the path to this transformative success, sparked by speech analysis using deep learning methods on spectrogram-like raw features and then progressing rapidly to speech recognition with increasingly larger vocabularies and scale. The role of well-timed academic-industrial collaboration will be highlighted, so will be the advances of big data, big compute, and the seamless integration between the application-domain knowledge of speech and general principles of deep learning. Then, an overview will be given on sweeping achievements of deep learning in speech recognition since its initial success in 2010 (as well as in image recognition and computer vision since 2012). Such achievements have resulted in across-the-board, industry-wide deployment of deep learning. The final part of the talk will look ahead towards stimulating new challenges of deep learning --- making intelligent machines capable of not only hearing (speech) and seeing (vision), but also of thinking with a “mind”; i.e. reasoning and inference over complex, hierarchical relationships and knowledge sources that comprise a vast number of entities and semantic concepts in the real world based in part on multi-sensory data from the user. To this end, language and multimodal processing --- joint exploitation and learning from text, speech/audio, and image/video --- is evolving into a new frontier of deep learning, beginning to be embraced by a mixture of research communities including speech and spoken language processing, natural language processing, computer vision, machine learning, information retrieval, cognitive science, artificial intelligence, and data/knowledge management. A review of recent published studies will be provided on deep learning applied to selected language and multimodal processing tasks, with a trace back to the relevant early connectionist modeling and neural network literature and with future directions in this new exciting deep learning frontier discussed and analyzed.