arXiv (Cornell University) · 2018 · 355 citations · 0 references
ComputingComputer EngineeringComputer ScienceEmbedded SystemsInternet Of ThingsMachine LearningConversational User InterfacesDeep LearningLanguage ModelingPattern RecognitionSpeech InteractionSpeech RecognitionAutomatic RecognitionData-to-text GenerationNatural Language InterfacesSpeech UnderstandingSpoken Language GenerationTest Data GenerationVoice User InterfacesVoice-controlled Interfaces
This paper presents the machine learning architecture of the Snips Voice\nPlatform, a software solution to perform Spoken Language Understanding on\nmicroprocessors typical of IoT devices. The embedded inference is fast and\naccurate while enforcing privacy by design, as no personal user data is ever\ncollected. Focusing on Automatic Speech Recognition and Natural Language\nUnderstanding, we detail our approach to training high-performance Machine\nLearning models that are small enough to run in real-time on small devices.\nAdditionally, we describe a data generation procedure that provides sufficient,\nhigh-quality training data without compromising user privacy.\n