Amazon Technologies, Inc.

Ming Sun

Chao Wang

Shiv Naga Prasad Vitaladevuni

Yixin Gao

Spyridon Matsoukas

Chieh-Chi Kao

Qingming Tang

Viktor Rozgic

James G Droppo

Anish N Shah

Chia-Jung Chang

Qin Zhang

Bjorn Hoffmeister

Arindam Mandal

Mathias Jourdain

Sumit Garg

Sankaran Panchapagesan

Thibaud Senechal

Cengiz Erbas

Rong Chen

Chungnam Chan

Jason Krone

Yuzong Liu

Varun Kumar Nagaraja

Aaron David Wirshba

Nipul Bharani

Jialiang Bao

Sheetal Laad

Harshavardhan Sundar

Yang Liu

Nikko Strom

Ryan Paul Thomas

Gengshen Fu

Aparna Khare

Minhua Wu

Björn Hoffmeister

Aaron Lee Mathers Challenner

Weiran Wang

Spyros Matsoukas

Thibaud Senechai

Srinivas Parthasarathy

Sripal Mehta

Andrew Thomas Bydlon

Steve Mark Lorusso

David Snyder

Venkata Naga Krishna Chaitanya Puvvada

Sumig Garg

James Garnet Droppo

A system may include a first acoustic event detection (AED) component configured to detect a predetermined set of acoustic events, and include a second AED component configured to detect custom acoustic events that a user configures a device to detect. The first and second AED components are configured to perform task-specific processing, and may receive as input the same acoustic feature data corresponding to audio data that potentially represents occurrence of one or more events. Based on processing by the first and second AED components, a device may output data indicating that one or more acoustic events occurred, where the acoustic events may be a predetermined acoustic event and/or a custom acoustic event.

Acoustic event detection

A system may be configured to detect custom acoustic events, where the system generates an acoustic event profile for the custom acoustic event based on a natural language description provided by a user and using an audio sample of the described acoustic event. For example, the user may describe the custom acoustic event as 'dog bark.' The system may ask the user questions to refine the description (e.g., dog breed, dog gender, age, etc.). Using an audio sample of the refined description, the system may then determine that audio captured in the user's environment is a potential sample of the custom acoustic event. Such captured audio may be presented to the user for confirmation, and then may be used to detect future occurrences of the custom acoustic event in the user's environment.

Techniques for detecting certain acoustic events from audio data are described. A system may perform event aggregation for certain types of events before sending an output to a device representing the event is detected. The system may bypass the event aggregation process for certain types of events that the system may detect with a high level of confidence. In such cases, the system may send an output to the device when the event is detected. The system may be used to detect acoustic events representing presence of a person or other harmful circumstances (such as, fire, smoke, etc.) in a home, an office, a store, or other types of indoor settings.

An acoustic event detection system may employ self-supervised federated learning to update encoder and/or classifier machine learning models. In an example operation, an encoder may be pre-trained to extract audio feature data from an audio signal. A decoder may be pre-trained to predict a subsequent portion of audio data (e.g., a subsequent frame of audio data represented by log filterbank energies). The encoder and decoder may be trained using self-supervised learning to improve the decoder's predictions and, by extension, the quality of the audio feature data generated by the encoder. The system may apply federated learning to share encoder updates across user devices. The system may fine-tune the classifier to improve inferences based on the improved audio feature data. The system may distribute classifier updates to the user device(s) to update the on-device classifier.

Self-supervised federated learning

A system and method performs multilingual wakeword detection by determining a language corresponding to the wakeword. A first wakeword-detection component, which may execute using a digital-signal processor, determines that audio data includes a representation of the wakeword and determines a language corresponding to the wakeword. A second, more accurate wakeword-detection component may then process the audio data using the language to confirm that it includes the representation of the wakeword. The audio data may then be sent to a remote system for further processing.

Multilingual wakeword detection

An acoustic event detection system may employ one or more recurrent neural networks (RNNs) to extract features from audio data, and use the extracted features to determine the presence of an acoustic event. The system may use self-attention to emphasize features extracted from portions of audio data that may include features more useful for detecting acoustic events. The system may perform self-attention in an iterative manner to reduce the amount of memory used to store hidden states of the RNN while processing successive portions of the audio data. The system may process the portions of the audio data using the RNN to generate a hidden state for each portion. The system may calculate an interim embedding for each hidden state. An interim embedding calculated for the last hidden state may be normalized to determine a final embedding representing features extracted from the input data by the RNN.

Streaming self-attention in a neural network

Described herein is a system for sentiment detection in audio data. The system processes audio frame level features of input audio data using a machine learning algorithm to classify the input audio data into a particular sentiment category. The machine learning algorithm may be a neural network trained using an encoder-decoder method. The training of the machine learning algorithm may include normalization techniques to avoid potential bias in the training data that may occur when the training data is annotated for a perceived sentiment of the speaker.

Sentiment detection in audio data

A system processes audio data to detect when it includes a representation of a wakeword or of an acoustic event. The system may receive or determine acoustic features for the audio data, such as log-filterbank energy (LFBE). The acoustic features may be used by a first, wakeword-detection model to detect the wakeword; the output of this model may be further processed using a softmax function, to smooth it, and to detect spikes. The same acoustic features may be also be used by a second, acoustic-event-detection model to detect the acoustic event; the output of this model may be further processed using a sigmoid function and a classifier. Another model may be used to extract additional features from the LFBE data; these additional features may be used by the other models.

Wakeword and acoustic event detection

A system and method performs wakeword detection and automatic speech recognition using the same acoustic model. A mapping engine maps phones/senones output by the acoustic model to phones/senones corresponding to the wakeword. A hidden Markov model (HMM) may determine that the wakeword is present in audio data; the HMM may have multiple paths for multiple wakewords or may have multiple models. Once the wakeword is detected, ASR is performed using the acoustic model.

Wakeword detection using multi-word model

A system may include an acoustic event detection component for detecting acoustic events, which may be non-speech sounds. Upon detection of a command to detect a new sound, a device may prompt a user to cause occurrence of the sound one or more times. The acoustic event detection component may then be reconfigured, using audio data corresponding to the occurrences, to detect future occurrences of the event.

Described herein is a system for media presence detection in audio. The system analyzes audio data to recognize whether a given audio segment contains sounds from a media source as a way of differentiating recorded media source sounds from other live sounds. In exemplary embodiments, the system includes a hierarchical model architecture for processing audio data segments, where individual audio data segments are processed by a trained machine learning model operating locally, and another trained machine learning model provides historical and contextual information to determine a score indicating the likelihood that the audio data segment contains sounds from a media source.

Media presence detection

Growing community of inventors

Winchester, MA, United States of America

Ming Sun