Google Inc.

Dharmeshkumar Mokani

Madhukar N Thakur

Nandini Kappiah

Morgan Avery Conbere

Andrew Granoff

Guangqi Ye

Francesco Xavier Llora Fabrega

Qi Cao

Karen Lees

Olawale Abiri

Oleg Chernyakhovskiy

Luis Alberto Rojas Marambio

A method for speech recognition biasing includes receiving, from an application executing on a user device, at a speech service interface, a speech recognition request requesting a transcription of an utterance. The speech recognition request includes audio data encoding the utterance and configuration parameters for biasing a speech recognition model based on context data. The method includes processing, using the speech recognition model, the audio data to generate speech recognition scores for speech elements and determining context scores for the speech elements based on the configuration parameters and the context data. The method includes biasing the speech recognition scores using the context scores. The method also includes determining the transcription for the utterance based on the biased speech recognition scores.

Speech recognition biasing

The present technology is directed to systems and methods for determining content of interest by determining the occurrence of fast-rising features (for example, URL, fingerprint, etc.) in posts or comments provided by users within an online community, for example, a social network. The systems and methods obtain user-generated content from an online community and determine signals based on the user-generated content to calculate a frequency of the signals as they post or occur in a pre-defined time bucket or window. A first derivative difference of the frequency and a second derivative difference of the frequency is calculated based on formulating consecutive time buckets or 'dynamic windows' to determine if one or more of the frequency, the first derivative difference, and the second derivative difference satisfies one or more pre-defined thresholds. The user-generated content is prioritized based on whether it satisfies the one or more pre-defined thresholds.

Determining content classifications using feature frequency

The present disclosure includes systems and methods for classifying user content (e.g., spam) within an online community by determining user specific signals and content specific signals, the systems and methods comprising: a user interface module configured to receive user content from the online community, a social signal processing module configured to determine signals from the user content, a user and content classifying module configured to classify user content and configured to compute a decision on the user content, a user content labeling module configured to label the user content, and an execution module configured to execute the decision.

Determining user content classifications within an online community

Growing community of inventors

San Jose, CA, United States of America

Dharmeshkumar Mokani