Google Inc.

Gang Feng

Xin Li

Cheng Xu

Fang Chu

Pedro Jose Moreno Mengibar

Ignacio Lopez Moreno

Quan Wang

Bo Zheng

Jin Shi

Jason William Pelecanos

Yeming Fang

Dylan Myers

A method of generating an accurate speaker representation for an audio sample includes receiving a first audio sample from a first speaker and a second audio sample from a second speaker. The method includes dividing a respective audio sample into a plurality of audio slices. The method also includes, based on the plurality of slices, generating a set of candidate acoustic embeddings where each candidate acoustic embedding includes a vector representation of acoustic features. The method further includes removing a subset of the candidate acoustic embeddings from the set of candidate acoustic embeddings. The method additionally includes generating an aggregate acoustic embedding from the remaining candidate acoustic embeddings in the set of candidate acoustic embeddings after removing the subset of the candidate acoustic embeddings.

Speaker identification accuracy

Business information about business entities are received from a plurality of aggregate information sources such as business directories. Mobile application logs about user activities are received from a plurality of mobile devices. Business entities related to the user activities are identified based on the mobile application logs. Scored attributes about the related business entities are acquired by applying data analysis rules to the mobile application logs. Accuracy scores are determined for attribute values in the business information based on the acquired scored attributes. Updated business information for the business entities is generated based on the accuracy scores and outputted to users upon request.

Using mobile application logs to measure and maintain accuracy of business information

Systems, methods, and computer readable storage mediums are provided to automatically identifying a classified website. A website is determined to be a candidate site based on a set of heuristics. From among pages constituting the candidate site one or more pages are determined to be listing page candidates and one or more pages are determined to be detail page candidates. Then a listing page score is determined using a listing page classifier. Similarly, a detail page score is determined using a detail page classifier. The listing page and detail page scores each indicate the likelihood that the pages are part of a classified website. A candidate site score is determined based in part on a combination of the listing page score and the detail page scores. Then when the candidate site score is above a threshold the candidate site is determined to be a classified website.

System and method for automatically identifying classified websites

Systems, methods, and computer readable storage mediums are provided for automatically extracting data from a classified website. A website is determined to be a classified website based on a set of heuristics. Then page models for other classified websites are accessed. The page models may include listing page models, detail page models, and/or city page models. A listing page in the classified website is determined based on similarity of the listing page to the page models for the other classified websites. Then a listing page model for the listing page in the classified website is created. After the model has been created data from the classified website is extracted based at least in part on the listing page model. Similar processes are performed for determining a details page, creating a details page model, and extracting data from the classified website using a details page model.

System and method for extracting structured data from classified websites

Growing community of inventors

Beijing, China

Gang Feng