Abstract: Methods, apparatuses and systems directed to summarizing video or other multimedia content based on the blogging activities of one or more users. In one implementation, a user may select a video and one or more blogs that annotate various segments of the video. A user may specify one or more desired attributes of the blog or comment entries, which a system processes to create an edit decision list. The edit decision list can be used to generate a summarized version of the multimedia content that includes the content segments associated with blog entries that meet the desired attributes.

Abstract: Embodiments of the invention are directed to using image data and contextual data to determine information about a scene, based on one or more previously obtained images. Contextual data, such location of image capture, can be used to determine previously obtained images related to the contextual data and other location-related information, such as billboard locations. With even low resolution devices, such as cell phone, image attributes, such as a histogram or optically recognized characters, can be compared between the previously obtained images and the newly captured image. Attributes matching within a predefined threshold indicate matching images. Information on the content of matching previously obtained images can be provided back to a user who captured the new image. User profile data can refine the content information. The content information can also be used as search terms for additional searching or other processing.

Abstract: A search engine determines a set of other images that are similar to a user-selected image, and presents those other images to the user. In determining whether two images are sufficiently similar to each other to merit presentation of one, the search engine determines a Euclidean distance between separate feature vectors that are associated with each of the images. Each such vector indicates diverse types of information that is known about the associated image. The types of information included within such a vector may include attributes that reflect visual characteristics that are visible in an image, verbal tags that have been associated with the image users in a community of users, concepts derived from those tags, coordinates that reflect a geographic location at which a camera that produced the image was when the camera produced the image, and concepts related to groups with which the image is associated.

Abstract: Pairs of feature vectors are obtained that represent speech. Some pairs represent two samples of speech from the same speakers, and other pairs represent two samples of speech from different speakers. A neural network feeds each feature vector in a sample pair into a separate bottleneck layer, with a weight matrix on the input of both vectors tied to one another. The neural network is trained using the feature vectors and an objective function that induces the network to classify whether the speech samples come from the same speaker. The weights from the tied weight matrix are extracted for use in generating derived features for a speech processing system that can benefit from features that are thus transformed to better reflect speaker identity.

Abstract: Pairs of feature vectors are obtained that represent speech. Some pairs represent two samples of speech from the same speakers, and other pairs represent two samples of speech from different speakers. A neural network feeds each feature vector in a sample pair into a separate bottleneck layer, with a weight matrix on the input of both vectors tied to one another. The neural network is trained using the feature vectors and an objective function that induces the network to classify whether the speech samples come from the same speaker. The weights from the tied weight matrix are extracted for use in generating derived features for a speech processing system that can benefit from features that are thus transformed to better reflect speaker identity.

Abstract: Improving accuracy in understanding and/or resolving references to visual elements in a visual context associated with a computerized conversational system is described. Techniques described herein leverage gaze input with gestures and/or speech input to improve spoken language understanding in computerized conversational systems. Leveraging gaze input and speech input improves spoken language understanding in conversational systems by improving the accuracy by which the system can resolve references—or interpret a user's intent—with respect to visual elements in a visual context. In at least one example, the techniques herein describe tracking gaze to generate gaze input, recognizing speech input, and extracting gaze features and lexical features from the user input. Based at least in part on the gaze features and lexical features, user utterances directed to visual elements in a visual context can be resolved.

Abstract: Systems and methods for generating and playing a sequence of media objects based on a mood gradient are also disclosed. A mood gradient is a sequence of items, in which each item is media object having known characteristics or a representative set of characteristics of a media object, that is created or used by a user for a specific purpose. Given a mood gradient, one or more new media objects are selected for each item in the mood gradient based on the characteristics associated with that item. In this way, a sequence of new media objects is created but the sequence exhibits a similar variation in media object characteristics. The mood gradient may be presented to a user or created via a display illustrating a three-dimensional space in which each dimension corresponds to a different characteristic. The mood gradient may be represented as a path through the three-dimensional space and icons representing media objects are located within the three-dimensional space based on their characteristics.

Abstract: Technologies described herein relate to modifying visual content for presentment on a display to facilitate improving performance of an automatic speech recognition (ASR) system. The visual content is modified to move elements further away from one another, wherein the moved elements give rise to ambiguity from the perspective of the ASR system. The visual content is modified to take into consideration accuracy of gaze tracking. When a user views an element in the modified visual content, the ASR system is customized as a function of the element being viewed by the user.

Abstract: A system and a corresponding method for temporal modification of audio signals, to increase or reduce the playback rates of an audio and/or a video file in a client-server environment. The system and method improve the efficiency of serving streaming media to a client so that the client can select an arbitrary time-speedup factor. The speedup system performs many of the pre-calculations once, at the server, so that the bandwidth needs are reduced and the client's computational load is minimized. The final time-scale-modification can be either done completely on the server, thus reducing the client's needs, or partly on the client's computer to minimize latency, and to reduce on-the-fly computational load from the server that serves multiple clients concurrently.

Abstract: Methods and system for identifying multimedia content streaming through a television includes retrieving an audio signal from a multimedia content selected for rendering at the television. The retrieved audio signal is partitioned into a plurality of segments of small intervals. A particular segment is analyzed to identify acoustic modulation and to generate a distinct vector for the particular segment based on the acoustic modulation, wherein the vector defines a unique fingerprint of the particular segment of the audio signal. A content database on a server is queried using the vector of the particular segment to obtain content information for multimedia content that matches the fingerprint of the particular segment. The content information is used to identify the multimedia content and the source of the multimedia content that matches the audio signal received for rendering.

Abstract: A method for displaying interactive advertisements on a television having a controller connected thereto and configured for receiving input from a viewer of the television is disclosed. The controller has a receiver operable to receive advertisements and a processor operable to modify the advertisements. The method generally comprises requesting action by the viewer of the television, modifying an advertisement based on the action of the viewer, and displaying the modified advertisement on the television.

Abstract: Software for initialized explore-exploit creates a plurality of probability distributions. Each of these probability distributions is generated by inputting a quantitative description of one or more features associated with an image into a regression model that outputs a probability distribution for a measure of engagingness for the image. Each of the images is conceptually related to the other images. The software uses the plurality of probability distributions to initialize a multi-armed bandit model that outputs a serving scheme for each of the images. Then the software serves a plurality of the images on a web page displaying search results, based at least in part on the serving scheme.

Abstract: The invention enables and/or facilitates the display of personalized content to content observers by content providers (e.g., advertisers) who disseminate content over a network. In particular, the invention can be used to enable the display of targeted content to television viewers and/or targeted product placements within a television program. The invention can advantageously enable identification of the content observer(s) observing a content display device at a particular time and/or identification of the content display site, and selection of targeted content for display based on a determination of content appropriate for display to those content observer(s) in accordance with a criterion based on the identity and/or a characteristic of the content observer(s) and/or the content display site.

Abstract: Systems and methods for generating and playing a sequence of media objects based on a mood gradient are also disclosed. A mood gradient is a sequence of items, in which each item is media object having known characteristics or a representative set of characteristics of a media object, that is created or used by a user for a specific purpose. Given a mood gradient, one or more new media objects are selected for each item in the mood gradient based on the characteristics associated with that item. In this way, a sequence of new media objects is created but the sequence exhibits a similar variation in media object characteristics. The mood gradient may be presented to a user or created via a display illustrating a three-dimensional space in which each dimension corresponds to a different characteristic. The mood gradient may be represented as a path through the three-dimensional space and icons representing media objects are located within the three-dimensional space based on their characteristics.

Abstract: Methods and apparatus, including computer program products, for recommendations based on adoption curve. A method includes tracking over a period of time user popularity of a media content using a web service residing in a server, the popularity and period of time representing a life cycle of each of the media content. The method tracks over the period of time users enrolled in the web service and when in the period of time each of the users adopted the tracked media content and associates adopted media content with user profiles representing the users. The method recommends media content associated with a first user who adopted the tracked media content earlier in the period of time to a second user who may want to adopt the tracked media content subsequently in the period of time.

Abstract: A method of checking whether a content aggregator's content matches a content owner's content involves generating a fingerprint of the content and looking for a matching fingerprint from the content owner through a service provided by the content owner. In one aspect, the fingerprints are generated from an intermediate digest of the content instead of the original form.

Abstract: Systems and methods for generating and playing a sequence of media objects based on a mood gradient are also disclosed. A mood gradient is a sequence of items, in which each item is media object haying known characteristics or a representative set of characteristics of a media object, that is created or used by a user for a specific purpose. Given a mood gradient, one or more new media objects are selected for each item in the mood gradient based on the characteristics associated with that item. In this way, a sequence of new media objects is created but the sequence exhibits a similar variation in media object characteristics. The mood gradient may be presented to a user or created via a display illustrating a three-dimensional space in which each dimension corresponds to a different characteristic. The mood gradient may be represented as a path through the three-dimensional space and icons representing media objects are located within the three-dimensional space based on their characteristics.

Abstract: Systems and methods for generating and playing a sequence of media objects based on a mood gradient are also disclosed. A mood gradient is a sequence of items, in which each item is media object having known characteristics or a representative set of characteristics of a media object, that is created or used by a user for a specific purpose. Given a mood gradient, one or more new media objects are selected for each item in the mood gradient based on the characteristics associated with that item. In this way, a sequence of new media objects is created but the sequence exhibits a similar variation in media object characteristics. The mood gradient may be presented to a user or created via a display illustrating a three-dimensional space in which each dimension corresponds to a different characteristic. The mood gradient may be represented as a path through the three-dimensional space and icons representing media objects are located within the three-dimensional space based on their characteristics.

Abstract: An online music system includes a music database configured to store musical selections and to store a user profile for respective users of the online music system, an advertiser account management system to store bid amounts from advertisers seeking to provide information to the users of the online music system and a user recommendation system coupled to the music database to present information about musical selections to respective users based on the stored user profile and the stored bid amounts.

Abstract: Software for initialized explore-exploit creates a plurality of probability distributions. Each of these probability distributions is generated by inputting a quantitative description of one or more features associated with an image into a regression model that outputs a probability distribution for a measure of engagingness for the image. Each of the images is conceptually related to the other images. The software uses the plurality of probability distributions to initialize a multi-armed bandit model that outputs a serving scheme for each of the images. Then the software serves a plurality of the images on a web page displaying search results, based at least in part on the serving scheme.