Soundhound, Inc.

Walmart Apollo, LLC

Kiran Garaga Lokeswarappa

Bernard Mont-Reynaud

Jun Huang

Joel Gedalius

Jonah Probell

Richard Edward Chatwin

Murthy V Nukala

Terry Kong

Alan Coleman

Steffen Holm

Srinidhi Ramesh Kondaji

A computer-implemented method is provided. The method including receiving speech audio of dictation associated with a user ID, deriving acoustic features from the speech audio, storing the derived acoustic features in a user profile associated with the user ID, receiving a request for acoustic features through an application programming interface (API), the request including the user ID, and sending the derived acoustic features through the API.

Deriving acoustic features and linguistic features from received speech audio

In order to improve the accuracy of ASR, an utterance is transcribed using a plurality of language models, such as for example, an N-gram language model and a neural language model. The language models are trained separately. They each output a probability score or other figure of merit for a partial transcription hypothesis. Model scores are interpolated to determine a hybrid score. While recognizing an utterance, interpolation weights are chosen or updated dynamically, in the specific context of processing. The weights are based on dynamic variables associated with the utterance, the partial transcription hypothesis, or other aspects of context.

Dynamic interpolation for hybrid language models

A method is provided for advertisement selection. The method includes recognizing words from user speech over a large number of interactions, computing a number of unique words uttered during the interactions, classifying the user by the number of unique words uttered during the interactions, and selecting an advertisement targeted to the classified users.

Advertisement selection by linguistic classification

Speech synthesis chooses pronunciations of words with multiple acceptable pronunciations based on an indication of a personal, class-based, or global preference or an intended non-preferred pronunciation. A speaker's words can be parroted back on personal devices using preferred pronunciations for accent training. Degrees of pronunciation error are computed and indicated to the user in a visual transcription or audibly as word emphasis in parroted speech. Systems can use sets of phonemes extended beyond those generally recognized for a language. Speakers are classified in order to choose specific phonetic dictionaries or adapt global ones. User profiles maintain lists of which pronunciations are preferred among ones acceptable for words with multiple recognized pronunciations. Systems use multiple correlations of word preferences across users to predict use preferences of unlisted words. Speaker-preferred pronunciations are used to weight the scores of transcription hypotheses based on phoneme sequence hypotheses in speech engines.

Pronunciation guided by automatic speech recognition

A system and method are provided for adding user characterization information to a user profile by analyzing user's speech. User properties such as age, gender, accent, and English proficiency may be inferred by extracting and deriving features from user speech, without the user having to configure such information manually. A feature extraction module that receives audio signals as input extracts acoustic, phonetic, textual, linguistic, and semantic features. The module may be a system component independent of any particular vertical application or may be embedded in an application that accepts voice input and performs natural language understanding. A profile generation module receives the features extracted by the feature extraction module and uses classifiers to determine user property values based on the extracted and derived features and store these values in a user profile. The resulting profile variables may be globally available to other applications.

Method and system for building an integrated user profile

In one embodiment, a method includes receiving, in real-time, user information associated with a user of a client computing device. One or more user features are derived from the user information associated with the user. The one or more user features include one or more user intents. The one or more user intents are matched with one or more of a plurality of advertiser intents. The user information is determined to be accepted. In response to accepting the user information, an advertisement is generated based on the matching of the one or more user intents and the one or more of the plurality of advertiser intents.

Growing community of inventors

Mountain View, CA, United States of America

Kiran Garaga Lokeswarappa