Amazon Technologies, Inc.

Dell Products Inc.

Parind Shah

Mohamed Farouk Mansour

Yuan-Yen Tai

Shiv Naga Prasad Vitaladevuni

Thibaud Senechal

Anish N Shah

Yuriy Mishchenko

Christin Jose

Alex Escott

A system and method performs wakeword detection using a feedforward neural network model. A first output of the model indicates when the wakeword appears on a right side of a first window of input audio data. A second output of the model indicates when the wakeword appears in the center of a second window of input audio data. A third output of the model indicates when the wakeword appears on a left side of a third window of input audio data. Using these outputs, the system and method determine a beginpoint and endpoint of the wakeword.

Wakeword detection using a neural network

A system may embed audio watermarks in audio data using a sign sequence. The system may detect audio watermarks in audio data despite the effects of reverberation. For example, the system may embed multiple repetitions of an audio watermark before generating output audio using loudspeaker(s). To detect the audio watermark in audio data generated by a microphone, the system may perform a self-correlation that indicates where the audio watermark is repeated. In some examples, the system may encode the audio watermark using multiple repetitions of a multi-segment Eigenvector. Additionally or alternatively, the system may encode the audio watermark using a binary sequence of positive and negative values, which may be used as a shared key for encoding/decoding the audio watermark. The audio watermark can be embedded in output audio data to enable wakeword suppression (e.g., avoid cross-talk between devices) and/or local signal transmission between devices in proximity to each other.

Audio watermark encoding/decoding

A system may embed audio watermarks in audio data using an Eigenvector matrix. The system may detect audio watermarks in audio data despite the effects of reverberation. For example, the system may embed multiple repetitions of an audio watermark before generating output audio using loudspeaker(s). To detect the audio watermark in audio data generated by a microphone, the system may perform a self-correlation that indicates where the audio watermark is repeated. In some examples, the system may encode the audio watermark using multiple repetitions of a multi-segment Eigenvector. Additionally or alternatively, the system may encode the audio watermark using a binary sequence of positive and negative values, which may be used as a shared key for encoding/decoding the audio watermark. The audio watermark can be embedded in output audio data to enable wakeword suppression (e.g., avoid cross-talk between devices) and/or local signal transmission between devices in proximity to each other.

A data handling system having a physical storage device and a storage controller responsible for provisioning, managing, and servicing logical unit numbers (LUNs) with self-destruction properties on the physical storage device is disclosed. For a given LUN, the storage controller creates a profile including self-destruction properties, such as a LUN destruction date, and associates the profile with the LUN. The profiles may be independent of file format and content of any associated data and the LUN destruction date may be a function of the last access date of the associated data. The storage controller monitors the LUN destruction date associated with each LUN and determines the last access date of any associated data. The storage controller marks a LUN having a past-due LUN destruction date, notifies a user of an approaching LUN destruction date, and destroys any past-due LUNs.

Growing community of inventors

Chelmsford, MA, United States of America

Parind Shah