Low-Complexity Speech Recognition for Next Generation Vocal User Interfaces

人工智能旨在改善加拿大人在医疗保健、信息技术、娱乐和工业自动化等许多领域的生活质量。机器学习的最新进展，特别是深度神经网络，已被证明为图像和语音处理中的许多挑战性问题提供了最知名的解决方案。特别是，总部位于蒙特利尔的研发公司Fluent.ai开发了一种深度学习算法，用于完全声学的语音界面-直接将语音数据转换为动作，消除了首先将语音转换为文本，然后将文本转换为动作的昂贵的两步过程。Fluent.ai希望麦吉尔团队研究在受限复杂度系统中直接语音到动作机器学习算法的低复杂度实现。该研究项目将用于协助Fluent.ai评估低复杂度神经网络在智能玩具和个人数字助理等产品中的适用性。

Artificial intelligence aims to improve the quality of lives of Canadians in many areas such as health care,information technology, entertainment and industrial automation. Recent advances in machine learning, inparticular deep neural networks, have been shown to provide the best-known solutions to many challengingproblems in image and speech processing. In particular, Fluent.ai, a Montreal-based R&D company, hasdeveloped a deep learning algorithm for an entirely acoustic voice interface - converting speech data to actionsdirectly, eliminating the costly two-step process of first converting speech to text and then converting text toaction. Fluent.ai would like the McGill team to investigate low-complexity implementation of directspeech-to-action machine learning algorithms in constrained-complexity systems. This research project will beused to assist Fluent.ai in evaluating the applicability of low-complexity neural networks for their customerneeds in products such as smart toys and personal digital assistants.