Abstract: The present disclosure relates to a speaker model adaptation method and device for enhancing text-independent speaker recognition performance. Specifically, the disclosure relates to a method and a device whereby, for the adaption of a speaker model pre-stored in an electronic device, text-independent speaker recognition performance is improved by considering variations in the amount of speaker characteristics information per phoneme unit.

Abstract: An approach for determining a veracity of a reported event is provided. In an embodiment, a set of predictor variables is retrieved from a selected use case. Each of these predictor values is a condition that indicates the veracity of the reported event. In addition, a set of hidden predictor variables is generated from a set of unstructured documents related to the reported event using a hidden Markov model that is based on the predictor variables using a cognitive system. These hidden predictor variables are combined with the set of predictor variables to generate a set of updated predictor variables. These updated predictor variables are used by the cognitive system to return a determination of the veracity of the reported event.

Abstract: Embodiments of a progressive feature selection method that selects features in multiple rounds are described. In one embodiment, the progressive feature selection method splits the feature space into tractable sub-spaces such that a feature selection algorithm can be performed on each sub-space. In a merge-split operation, the subset of features that the feature selection algorithm selects from the different sub-spaces are merged into subsequent sets of features. Instead of re-generating the mapping table for each subsequent set from scratch, a new mapping table from the previous round's tables is created by collecting those entries that correspond to the selected features. The feature selection method is then performed again on each of the subsequent feature sets and new features are selected from each of these feature sets. This feature selection-merge-split process is repeated on successively smaller numbers of feature sets until a single final set of features is selected.

Abstract: Embodiments of a feature generation system and process for use in machine learning applications utilizing statistical modeling systems are described. In one embodiment, the feature generation process generates large feature spaces by combining features using logical, arithmetic and/or functional operations. A first set of features in an initial feature space are defined. Some or all of the first set of features are processed using one or more arithmetic, logic, user-defined combinatorial processes, or combinations thereof, to produce additional features. The additional features and at least some of the first set of features are combined to produce an expanded feature space. The expanded feature space is processed through a feature selection and optimization process to produce a model in a statistical modeling system.

Abstract: A system and method recognizes speech securely. The system includes a client and a server, The client is configured to provide securely speech in a form of an observation sequence of symbols, and the server is configured to provide securely a multiple trained hidden Markov models (HMMs), each trained HMM including a multiple states, a state transition probability distribution and an initial state distribution, and each state including a subset of the observation symbols and an observation symbol probability distribution. The observation symbol probability distributions are modeled by mixtures of Gaussian distributions. Also included are means for determining securely, for each HMM, a likelihood the observation sequence is produced by the states of the HMM, and means for determining a particular symbol with a maximum likelihood of a particular subset of the symbols corresponding to the speech.