Amazon Technologies, Inc.

Jeffrey Paul Lilly

Jeffrey Penrod Adams

Ryan Paul Thomas

Alok Ulhas Parlikar

Ariya Rastrow

Stan Weidner Salvador

Jeffrey C O'Neill

Frederick Victor Weber

Thomas Schaaf

Andrew Jake Rosenbaum

An automatic speech recognition (ASR) device may be configured to predict pronunciations of textual identifiers (for example, song names, etc.) based on predicting one or more languages of origin of the textual identifier. The one or more languages of origin may be determined based on the textual identifier. The pronunciations may include a hybrid pronunciation including a pronunciation in one language, a pronunciation in a second language and a hybrid pronunciation that combines multiple languages. The pronunciations may be added to a lexicon and matched to the content item (e.g., song) and/or textual identifier. The ASR device may receive a spoken utterance from a user requesting the ASR device to access the content item. The ASR device determines whether the spoken utterance matches one of the pronunciations of the content item in the lexicon. The ASR device then accesses the content when the spoken utterance matches one of the potential textual identifier pronunciations.

Predicting pronunciation in speech recognition

An automatic speech recognition (ASR) system may add new words to an ASR system by identifying words with similar usage and replicating the variations of the identified words to create new words. A new word that is used similarly to a known word may be varied to create new word forms that are similar to the word forms of a known word. The new word forms may then be incorporated into an ASR model to allow the ASR system to recognize those words when they are detected in speech. Such a system may allow flexible incorporation and recognition of varied forms of new words entering a general lexicon.

Word generation for speech recognition

Features are disclosed for reducing errors in speech recognition processing. Methods for reducing errors can include receiving multiple speech recognition hypotheses based on an utterance indicative of a command or query of a user and determining a command or query within a grammar having a least amount of difference from one of the speech recognition hypotheses. The determination of the least amount of difference may be based at least in part on a comparison of individual subword units along at least some of the sequence paths of the speech recognition hypotheses and the grammar. For example, the comparison may be performed on the phoneme level instead of the word level.

Error reduction in speech processing

Features are disclosed for active learning to identify the words which are likely to improve the guessing and automatic speech recognition (ASR) after manual annotation. When a speech recognition system needs pronunciations for words, a lexicon is typically used. For unknown words, pronunciation-guessing (G2P) may be included to provide pronunciations in an unattended (e.g., automatic) fashion. However, having manually (e.g., by a human) annotated pronunciations provides better ASR than having automatic pronunciations that may, in some instances, be wrong. The included active learning features help to direct these limited annotation resources.

Active learning for lexical annotations

Features are provided for selectively scoring portions of user utterances based at least on articulatory features of the portions. One or more articulatory features of a portion of a user utterance can be determined. Acoustic models or subsets of individual acoustic model components (e.g., Gaussians or Gaussian mixture models) can be selected based on the articulatory features of the portion. The portion can then be scored using a selected acoustic model or subset of acoustic model components. The process may be repeated for the multiple portions of the utterance, and speech recognition results can be generated from the scored portions.

Selective speech recognition scoring using articulatory features

Natural language controlled devices may be configured to activate command recognition in response to one or more wake words. Techniques are provided to receive a candidate word for evaluation as a wake word that activates a natural language control functionality of a computing device. The candidate word may include one or more words or sounds. Values for multiple wake word metrics are then determined. The candidate word is evaluated based on the various wake word metrics.

Growing community of inventors

Seattle, WA, United States of America

Jeffrey Paul Lilly