Abstract: A method including training a plurality of learning systems, each learning system implementing a learning function and having an input and producing an output, initializing one or more data structures, and evaluating a target sample is described. Also described are methods that include initializing one or more data structures and evaluating a target sample for a best match.

Abstract: A set of reference videos is indexed to a reference index in order to facilitate matching of video content. An indexing module receives a set of reference fingerprints representing a set of reference videos and identifies keys contained in the reference fingerprints. Reference identifiers for the reference videos are stored in bins of the reference index associated with the identified keys. The bins in the reference index are sub-sampled to limit the number of reference identifiers stored in a given bin.

Abstract: Methods, systems, and computer program products are provided for mapping keywords to geographic features. One example method includes identifying location keywords for each of a multitude of granular locations, determining a feature size for grouping granular locations over an area of interest, determining geo data for one or more features, locating all granular locations which are associated with a given feature using the geo data and forming a set of granular locations per feature, aggregating the location keywords for each granular location in a set forming a keyword mapping for the given feature, receiving an indication of a geographic location that is proximate to a user or is of interest to the user, determining a geographic feature associated with the geographic location, and targeting content for delivery to the user based at least in part on the keyword mapping.

Abstract: Systems, methods, apparatuses, user interfaces and computer program products provide social and interactive applications for mass media based on real-time ambient-audio and/or video identification.

Abstract: Compression of an image is performed based on prediction of target blocks of an image from candidate source blocks of the image. Heuristics are used for identifying the candidate source blocks, for example, source blocks are selected from within a cluster of similar blocks obtained by K-means clustering. For each target block, a region adjacent to the target block is identified and a set of candidate source blocks along with candidate source regions adjacent to the candidate source blocks are identified. The candidate source regions are ranked based on the differences between the candidate source regions and the target source region. Each candidate source block is described using its rank and residual information describing differences between the candidate source block and the target block. The candidate source block that can be described using a minimum amount of information is selected for predicting the target block.

Abstract: Embodiments of the present invention recite a method and system for improving the fidelity of a dialog system. In one embodiment, a first input generated by a user of a first system operating in a first modality is accessed. In embodiments of the present invention, the first system also generates a first output corresponding to the first input. An second input from a second user, who is engaged in a conversation with the first user, is accessed by a second system. The second input is then utilized to modify the first output of the first system.

Abstract: A computer-implemented method is disclosed for generating a signature representing an input bit vector. A signature generator generates a primary min-hash value based on a primary permutation from a sequence of permutation blocks. If the primary min-hash value is lower than a threshold value, a secondary min-hash value is generated based on a secondary permutation from the same permutation block. The signature generator then determines one or more signature values based on the primary min-hash value, the secondary min-hash value or both. The one or more signature values are stored as elements of the signature.

Abstract: Systems, methods, devices, and computer program products provide social and interactive applications for mass media based on real time ambient-audio and/or video identification. In some implementations, a method includes: receiving descriptors identifying ambient audio associated with a media broadcast; comparing the descriptors to one or more reference descriptors; and determining a rating for the media broadcast based at least in part on the results of the comparison.

Abstract: A method is described that includes producing an audio spectrogram from a target sample, generating a number of fingerprints based on the audio spectrogram, comparing the series of fingerprints to samples in a data repository using wavelet coefficients, and identifying the target sample based on the matches found in the data repository.

Abstract: A set of reference videos is indexed to a reference index in order to facilitate matching of video content. An indexing module receives a set of reference fingerprints representing a set of reference videos and identifies keys contained in the reference fingerprints. Reference identifiers for the reference videos are stored in bins of the reference index associated with the identified keys. The bins in the reference index are sub-sampled to limit the number of reference identifiers stored in a given bin.

Abstract: Conditioned data is generated from multi-dimensional data populations. The multi-dimensional data populations include time-based media data, such as video data, or data associated with time-based media. The conditioned data is scaled such that dimensions of greater distinguishability are emphasized. To generate the conditioned data, a data conditioner receives populations of multi-dimensional data that have the same dimensions in the data. Data of a dimension is scaled based on a distinguishability factor calculated for the dimension. The distinguishability factor is calculated based on distributions in the dimension for the data populations. The data conditioning approach is scalable to any number of data populations and any number of data dimensions.

Abstract: A system varies the apparent display rate of an audiovisual display by modifying an original set of audio data in accordance with a target display rate, then modifying a related original set of video data to conform to the modifications made to the audio data set, such that the modified audio and video data sets are synchronized. The apparent display rate can be faster or slower than a normal display rate for the original sets of audio and video data. Preferably, the method for modifying the original audio data set is one that produces a modified audio data set that can be used to generate an audio display having little or no distortion.

Abstract: In general, the subject matter described in this specification can be embodied in methods, systems, and program products. A plurality of electronic training images that are each classified as displaying substantially pictures is obtained. A plurality of local image features in each of the plurality of electronic training images is identified. A plurality of weak classifiers are recursively applied to the local image features. During each iteration a weak classifier that accurately classifies the local images features is selected. After each selection of a weak classifier features that were misclassified by the selected weak classifier are given greater weight than features that were classified correctly by the selected weak classifier. For each selected weak classifier a hillclimbing algorithm is performed to attempt to improve the weak classifier. A strong classifier that is a weighted combination of the selected weak classifiers on which hillclimbing algorithms have been performed is produced.

Abstract: A system and method detects matches between portions of video content. A matching module receives an input video fingerprint representing an input video and a set of reference fingerprints representing reference videos in a reference database. The matching module compares the reference fingerprints and input fingerprints to generate a list of candidate segments from the reference video set. Each candidate segment comprises a time-localized portion of a reference video that potentially matches the input video. A classifier is applied to each of the candidate segments to classify the segment as a matching segment or a non-matching segment. A result is then outputted identifying a matching portion of a reference video from the reference video set based on the segments classified as matches.

Abstract: The invention enables the apparent display rate of an audiovisual display to be varied. The invention can modify an original set of audio data in accordance with a target display rate, then modify a related original set of video data to conform to the modifications made to the audio data set, such that the modified audio and video data sets are synchronized. When the modified audio and video data sets so produced are used to generate an audiovisual display, the audiovisual display has an apparent display rate that approximates the target display rate. The target display rate can be faster or slower than a normal display rate at which an audiovisual display system generates an audiovisual display from the original sets of audio and video data. The target display rate can be established solely by a user instruction, by analysis of the audiovisual data, or by modification of a user-specified nominal target display rate based upon analysis of the audiovisual data.

Abstract: Methods, systems, and apparatus, including computer program products, for ranking images are disclosed. An image search subsystem generates an adjustment factor representative of a quality measure of an image relative to a search query. The quality represents a relevance of the image to the query. The adjustment factor can be computed based on relevance data for the image to the query and image similarity data representing a relative similarity between the image and other images relevant to the query. The relevance data can be based on user actions in response to the image being included in search results for the query. The adjustment factor can be scaled based on whether the relevance data and the image similarity data both indicate that the image is relevant to the search query. A relevance score is computed based on the adjustment factor (e.g., a product of the adjustment factor and relevance score).

Abstract: Videos are received and indexed based on fingerprints generated for the videos, using lookup keys which correspond to sub portions of the reference fingerprints. Specifically, a set of lookup keys is selected such that the number of reference fingerprints such that the clumping associated with the associated each of the lookup keys is minimized. Clumping occurs when a number of reference identifiers associated with a lookup key exceeds a maximum value.

Abstract: Systems, methods, devices, and computer program products provide social and interactive applications for detecting repeating content in broadcast media. In some implementations, a method includes: generating a database of audio statistics from content; generating a query from the database of audio statistics; running the query against the database of audio statistics to determine a non-identity match; if a non-identity match exists, identifying the content corresponding to the matched query as repeating content.

Abstract: A set of reference videos is indexed to a reference index in order to facilitate matching of video content. An indexing module receives a set of reference fingerprints representing a set of reference videos and identifies keys contained in the reference fingerprints. Reference identifiers for the reference videos are stored in bins of the reference index associated with the identified keys. The bins in the reference index are sub-sampled to limit the number of reference identifiers stored in a given bin.

Abstract: A method including training a plurality of learning systems, each learning system implementing a learning function and having an input and producing an output, initializing one or more data structures, and evaluating a target sample is described. Also described are methods that include initializing one or more data structures and evaluating a target sample for a best match.