Google Inc.

Russell J Ryan

Matthew K Gray

Namrata U Godbole

Andrew Stephen Tomkins

Quoc V Le

Yonghui Wu

Ron J Weiss

Zhifeng Chen

Navdeep Jaitly

Ioannis Agiomyrgiannakis

Ryan Michael Rifkin

Samuel Bengio

Robert Andrew James Clark

R Michael Angelo

Roni Fischman Zeiger

Yuxuan Wang

Zongheng Yang

Ying Xiao

Thomas A Lasko

Nathan Dickerson Beach

Thomas W Colthurst

Marius G Andrei

Robert Andrew Clark

Ryan M Rifkin

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating speech from text. One of the systems includes one or more computers and one or more storage devices storing instructions that when executed by one or more computers cause the one or more computers to implement: a sequence-to-sequence recurrent neural network configured to: receive a sequence of characters in a particular natural language, and process the sequence of characters to generate a spectrogram of a verbal utterance of the sequence of characters in the particular natural language; and a subsystem configured to: receive the sequence of characters in the particular natural language, and provide the sequence of characters as input to the sequence-to-sequence recurrent neural network to obtain as output the spectrogram of the verbal utterance of the sequence of characters in the particular natural language.

End-to-end text-to-speech conversion

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for identifying an image for an entity. In one aspect, a method includes identifying a set of resources. Each resource can include at least one image and reference at least one entity. Each image is assigned to a particular entity based on an overall entity scores for the image relative to the entities. The overall entity score for an image and an entity can specify a measure of topical relatedness between the image and the entity. For each individual entity referenced by at least one of the resources, a group of images that have been assigned to the individual entity is identified. An image evocative of each individual entity is selected from the group based on image rank scores. The image rank score for an image can be determined based on visual characteristics of the image.

Identifying an image for an entity

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying entities using search results. One of the methods includes determining that a first search query includes a respective text reference to each of one or more predetermined attributes, wherein each attribute is associated with a first entity type; obtaining search results for the first search query from a search engine, the search results identifying a plurality of resources; identifying entities of the first entity type that are related to any of the plurality of resources identified by the search results; and selecting names of one or more of the identified entities of the first entity type to include in a response to the first search query.

Identifying entities using search results

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying entities using search results. One of the methods includes determining that a search query includes a respective text reference to each of one or more predetermined attributes, wherein each attribute is associated with an entity type; obtaining search results for the search query from a search engine, each search result identifying a respective resource; for each of a plurality of the obtained search results, determining an initial score for each of a plurality of entities of the entity type based on occurrences of names of the entity in the resource identified by the search result; generating a final score for each of the entities based on the initial scores; and selecting one or more names of entities of the entity type to include in a response to the search query based on the final scores.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying entities using search results. One of the methods includes the actions of determining that a first search query includes a respective text reference to each of one or more predetermined attributes, wherein each attribute is associated with a first entity type; for each of a plurality of entities of the first entity type, generating a combined search query that includes the first search query and a name of the entity; obtaining search results for each of the plurality of entities using the combined search query for each respective entity; and using the obtained search results to generate combined search results to include in a response to the first search query.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying entities using search results. One of the methods includes the actions of determining that a first search query includes a respective text reference to each of one or more predetermined attributes, wherein each attribute is associated with a first entity type; generating a second search query including the first search query and one or more terms that refer to the first entity type; and evaluating search results obtained for the second search query to select one or more names of entities of the first entity type to include in a response to the first search query.

Growing community of inventors

Cambridge, MA, United States of America

Russell J Ryan