Non-stationary Signal Feature Extraction and Analysis

信号处理继续在与语音、多媒体、医疗保健、国防、安全、电信、互联网、通信和能源系统相关的许多技术创新和进步中发挥基础性作用。在过去的15年里，瑞尔森大学的信号分析研究(SAR)小组致力于开发各种创新的技术和算法，用于处理和分析语音、音频、多媒体和生物医学信号。这些系统所涉及的信号的基本特征是它们复杂，通常持续时间长，难以解释，并且具有时变特性。为了从这些信号中提取有价值的信息(特征)并表征感兴趣的事件，并自动分类模式，需要设计复杂的信号分析算法(和分析工具)。拟议的NSERC Discovery Grant研究将系统地研究和开发数学方法、算法和工具，将一维(1D)信号映射到更高的维度，以便在多个层次自动提取信号特征，否则从传统技术中很难或不可能提取这些特征。它设想将信号的数学变换到更高的维度，随后的特征提取算法将揭示潜在的信号生成/修改机制，这些机制可能有助于识别隐藏/细微的签名，以便更好地识别和分类应用。提取的信号特征将进一步与适当的机器学习算法相结合，以提供增强且稳健的识别和分类性能效率。自动特征提取与分析有着广泛的实际应用，是语音、音频、多媒体、生物特征识别等智能系统中日常应用的基础。这些算法将应用于我们实验室收集的真实世界数据集和其他开源数据库。这些算法和数据库还将与其他感兴趣的研究小组共享，以利于其特定的应用领域(例如，能源或卫生部门的大数据分析)。该研究计划还将培训大量高素质的人员，他们最终将在对加拿大社会福祉和经济繁荣至关重要的各个行业和研究部门引领技术进步。

Signal processing continues to play a fundamental role in many technological innovations and advancements related to speech, multimedia, healthcare, defense, security, telecommunications, Internet,  and energy systems. For the past 15 years, the Signal Analysis Research (SAR) Group at Ryerson University is involved in developing various innovative techniques and algorithms for processing and analysis of speech, audio, multimedia and biomedical signals. The underlying characteristics of signals involved with these systems is that they are complex, typically long duration, difficult to interpret, and have time-varying properties. In order to extract valuable information (features) from these signals and characterize events of interest, and to automatically classify patterns, sophisticated signal analysis algorithms (and analytical tools) need to be designed. The proposed NSERC Discovery Grant research will systematically investigate and develop mathematical methods, algorithms and tools to map 1-dimensional (1D) signals into higher dimensions for automatically extracting signal features at multiple levels, which are otherwise difficult or impossible to extract from conventional techniques. It is envisioned the mathematical transformation of signals to higher dimensions and the subsequent feature extraction algorithms will reveal underlying signal generation/modification mechanisms that could be useful in recognizing hidden/subtle signatures for better recognition and classification applications. The extracted signal features will be further coupled with appropriate machine learning algorithms in providing enhanced and robust recognition and classification performance efficiencies. Automatic feature extraction and analysis has lots of practical applications, and is the foundation of everyday systems encountered in speech, audio, multimedia,  biometrics and many other intelligent systems. The algorithms will be applied to real world datasets collected in our lab and other open source databases. The algorithms and the databases will also be shared with other interested research groups for the benefit of their specific domain of application (e.g., big data analytics in energy or health sector). The research program will also train a large number of highly qualified personnel who could eventually lead technological advancement in various industry and research sectors that are crucial for the societal well-being and economic prosperity of Canada.**