Microsoft Technology Licensing, LLC

Li Jiang

Xuedong D Huang

Amit Kumar

William Ramsey

Mei-Yuh Hwang

Peter Bailey

Sandy Wong

Shajib Sadhukha

Michael Van Snellenberg

Paul Reinholdtsen

Qiang Wu

Paul A Viola

Milind V Mahajan

Zhaowei Charlie Jiang

Hsiao-Wuen Hon

Sanaz Ahari

Steven William Macbeth

Michael J Rozak

Michael William O'Connor

Shipeng Li

Jie Jerry Liu

Zheng Chen

Dong Li

Yun-Cheng Ju

Ye-Yi Wang

Xiaodong He

Julian James Odell

Shuangyu Chang

Dimitrios Lymberopoulos

Fileno A Alleva

Jian-Tao Sun

Wei Zhang

Xiaolong Li

Zhimin Zhang

Xiaochuan Ni

Hemant Malhotra

Andrew Paul McGovern

Rachel Imogen Morton

Gregory T Buehrer

Steve Macbeth

Jakub Jan Szymanski

Ivo Jose Garcia Dos Santos

Embodiments are provided for securing data access to machine learning training data at a plurality of distributed computing devices. Electronic content including original data that corresponds to a preferred data security level is divided into a plurality of microsegments. The plurality of microsegments is restrictively distributed to a plurality of computing devices which apply transcription labels to the plurality of microsegments. The labeled microsegments are reconstructed into training data which is then used to train a machine learning model while facilitating an improvement in data security of the original data included with the training data from the reconstructed microsegments.

Microsegment secure speech transcription

Among other things, one or more techniques and/or systems are provided for facilitating the completion of a user task. That is, user intent (e.g., intentions of a user to perform a user task) may be identified. The user intent may comprise an entity (e.g., a movie entity) and/or an action (e.g., an order movie tickets action) that the user wants to perform on the entity. A provider list may be created based upon one or more providers capable of performing the action on the entity (e.g., a movie application may be capable of performing the order movie tickets action on the movie entity). Providers may be dynamically selected for inclusion within the provider list at run-time. For example, an open market of providers may be maintained (e.g., providers may be added, removed, and/or updated over time), such that providers may be selected from the open market to complete user tasks.

User task completion via open market of actions and/or providers

Systems and methods are provided for performing tasks and related tasks. While a user is performing a task, a group of related tasks is presented. When a new task is selected, a new group of related tasks is provided. This allows a user to perform a series of tasks without having to separately search for each task. As the user moves between tasks, task information items are passed between the tasks so that previously provided information can be preserved.

Related tasks and tasklets for search

An application ('app') may be associated with a web site, so that an online service may use the connection between the app and the web site. In one example, the online service is an information provider, such as a search engine or advertising engine. The online service provides a web administrator portal, through which a webmaster provides or generates a secret code. The webmaster gives the secret code to an app developer, who gives the code to an app store when the app is published. If the service provides information about the web site (e.g., as a search result or ad impression), the service may also provide an indication of the app, since the use of the code represents confirmation by both the webmaster and app developer that a relationship exists between the web site and the app.

Connection between app and web site

A method includes receiving a search query from a user at a search engine. The method also includes performing a search responsive to the search query to obtain a set of results. The set of results is ordered at least partially based on a distance between a profile of a user and a profile of each result in the ordered set of results. At least one profile of a result is based on an aggregate of a plurality of user profiles.

Search results based on user and result profiles

Embodiments of the present invention relate to systems, methods, and computer-storage media for a method of contextually analyzing terms within a search query. In one embodiment, a received search query is classified into a domain category. Additionally, information is assigned to a schema associated with the domain by analyzing the search query. Further, at least one search result that helps a user complete a task within the domain is provided based on the information in the schema.

Dialog-enhanced contextual search query analysis

Methods, systems, and media are provided for facilitating generation of an inference index. In embodiments, a canonical entity is referenced. The canonical entity is associated with web documents. One or more queries that, when input, result in a selection of at least one of the web documents are identified. An entity document is generated for the canonical entity. The entity document includes the identified queries and/or associated text from the content of a document or from an entity title that result in the selection of the at least one of the web documents. The entity document and corresponding canonical entity can be combined with additional related entity documents and canonical entities to generate an inference index.

Inference indexing

Computer-readable media, computer systems, and computing methods are provided for implicitly discovering applications using a query or search results, where the query does not explicitly target a particular application. A search engine is used to receive a user-initiated query and to employ a keyword of the query to generate an initial set of search results. Characteristics of the search results are identified and applications that are relevant to the characteristics are determined. The determination of relevant applications include: accessing an application manifest that manages a mapping between applications and predefined characteristics; comparing the characteristics of the search results against the predefined characteristics of the application manifest to determine a match; and, when a match exists between the search-result characteristics and the predefined characteristics, identifying applications mapped to the matched characteristics as being relevant to the search results. These identified applications are presented in response to the query.

Integrating applications within search results

An expected dialog-turn (ED) value is estimated for evaluating a speech application. Parameters such as a confidence threshold setting can be adjusted based on the expected dialog-turn value. In a particular example, recognition results and corresponding confidence scores are used to estimate the expected dialog-turn value. The recognition results can be associated with a possible outcome for the speech application and a cost for the possible outcome can be used to estimate the expected dialog-turn value.

Confidence threshold tuning

A method and apparatus are provided for segmenting words into component parts. Under the invention, mutual information scores for pairs of graphoneme units found in a set of words are determined. Each graphoneme unit includes at least one letter. The graphoneme units of one pair of graphoneme units are combined based on the mutual information score. This forms a new graphoneme unit. Under one aspect of the invention, a syllable n-gram model is trained based on words that have been segmented into syllables using mutual information. The syllable n-gram model is used to segment a phonetic representation of a new word into syllables. Similarly, an inventory of morphemes is formed using mutual information and a morpheme n-gram is trained that can be used to segment a new word into a sequence of morphemes.

Generating large units of graphonemes with mutual information criterion for letter to sound conversion

A method and apparatus are provided that perform text normalization and inverse text normalization using a single grammar. During text normalization, a finite state transducer identifies a second string of symbols from a first string of symbols it receives. During inverse text normalization, the context free transducer identifies the first string of symbols after receiving the second string of symbols.

Method and apparatus for transducer-based text normalization and inverse text normalization

A method and apparatus is provided for using multiple feature streams in speech recognition. In the method and apparatus, a feature extractor generates at least two feature vectors for a segment of an input signal. A decoder then generates a path score that is indicative of the probability that a word is represented by the input signal. The path score is generated by selecting the best feature vector to use for each segment. For each segment, the corresponding part in the path score for that segment is based in part on a chosen segment score that is selected from a group of at least two segment scores. The segment scores each represent a separate probability that a particular segment unit (e.g. senone, phoneme, diphone, triphone, or word) appears in that segment of the input signal. Although each segment score in the group relates to the same segment unit, the scores are based on different feature vectors for the segment.

Speech recognition method and apparatus utilizing multiple feature streams

Growing community of inventors

Bellevue, WA, United States of America

Li Jiang