Amazon Technologies, Inc.

Francois Mairesse

Shiv Naga Prasad Vitaladevuni

Edward C Bueche

Paul Frederick Raccuglia

Simon Peter Reavely

Tao Ye

Amina Shabbeer

Yun Suk Paik

Jonathan Pollack

Charles Lee Thorp

Warren D Freitag

Anmol Tiwari

Amaury Gutierrez Acosta

Torbjorn Vik

Dialog acts (e.g., questions) are selected for voice browsing by a model trained to identify a dialog act that is most likely to lead to a desired outcome. Upon receiving an invocation to begin a conversation, a score indicative of a level of confidence that the conversation will have a successful outcome is determined, and a dialog act is selected based on the score. Subsequently, at each turn of the conversation, the score is updated or a new score is calculated, and a dialog act is selected based on the updated or new score. Confidence scores are calculated based on input features that are determined based on the user who uttered the invocation or responses to dialog acts, as well as a context of the conversation, and provided to a linear model or a machine learning model as inputs.

Confidence scoring for selecting tones and text of voice browsing conversations

Voice data from a current conversation between a user and a voice-controlled user device can be used to determine a search constraint for searching a database. Other search constraints can be determined based at least in part on the current conversation, a previous conversation, and/or a previous action. Properties can be associated with the search constraints. Once the search constraints have been determined, a plurality of search query plans is determined and a first search query plan is executed to query the database.

Voice conversational search constraint management

Dialog acts (e.g., questions) are selected for voice browsing by a machine learning model trained to identify a dialog act that is most likely to lead to a desired outcome. When an invocation is received from a user, a context of the invocation is determined, and a pool of dialog acts is scored based on the context by a machine learning model. Dialog acts are selected from the pool and presented to the user in accordance with a randomization policy. Data regarding the dialog acts and their success in achieving a desired outcome is used to train one or more machine learning models to select dialog acts in response to invocations.

Selecting dialog acts using controlled randomness and offline optimization

Automatic speech recognition (ASR) processing including a feedback configuration to allow for improved disambiguation between ASR hypotheses. After ASR processing of an incoming utterance where the ASR outputs an N-best list including multiple hypotheses, the multiple hypotheses are passed downstream for further processing. The downstream further processing may include natural language understanding (NLU) or other processing to determine a command result for each hypothesis. The command results are compared to determine if any hypotheses of the N-best list would yield similar command results. If so, the hypothesis(es) with similar results are removed from the N-best list so that only one hypothesis of the similar results remains in the N-best list. The remaining non-similar hypotheses are sent for disambiguation, or, if only one hypothesis remains, it is sent for execution.

Disambiguation in speech recognition

Automatic speech recognition (ASR) processing including a two-stage configuration. After ASR processing of an incoming utterance where the ASR outputs an N-best list including multiple hypotheses, a first stage determines whether to execute a command associated with one of the hypotheses or whether to output some of the hypotheses of the N-best list for disambiguation. A second stage determines what hypotheses should be included in the disambiguation choices. A first machine learning model is used at the first stage and a second machine learning model is used at the second stage. The multi-stage configuration allows for reduced speech processing errors as well as a reduced number of utterances sent for disambiguation, which thus improves the user experience.

Growing community of inventors

Arlington, MA, United States of America

Francois Mairesse