Acoustic event sensor
AssigneeApple Inc.
InventorsRobert D. Silvfast, Nicholas C. Soldner, Siddharth S. Hazra
Life of this patent
Estimating the term…
Abstract
An integrated circuit package having integrated therein an array of acoustic to electrical transducers that produce analog audio sub-band signals, an analog detector circuit to produce for each sub-band a number of analog property signals, an analog to digital conversion circuit to digitize each of the analog property signals, and an encoder to produce for each sub-band a feature set that contains one or more features of a respective sub-band. Other aspects are also described and claimed.
Claims19
A method for acoustic event sensing, the method comprising:
- separating acoustic energy into a plurality of sub-bands as a plurality of analog audio sub-band signals;
- extracting in analog form a plurality of properties from each of the plurality of analog audio sub-band signals wherein the plurality of properties include a first derivative and an envelope of the analog audio sub-band signal;
- digitizing the plurality of properties for each of the plurality of sub-bands, after extracting the plurality of properties in analog form;
- encoding the digitized plurality of properties for each of the sub-bands into a feature set being one or more features;
- and
- analyzing the feature set to classify the acoustic energy or detect a voice trigger phrase.
An apparatus for acoustic event sensing, comprising:
- an acoustic event sensor configured to separate acoustic energy into a plurality of sub-bands as a plurality of analog audio sub-band signals, extract in analog form a plurality of properties from each of the plurality of analog audio sub-band signals wherein the plurality of properties include a first derivative and an envelope of the analog audio sub-band signal, digitize the plurality of properties, and encode the digitized plurality of properties for each of the plurality of sub-bands into a feature set being one or more features;
- and
- a processor configured to analyze the feature set to classify the acoustic energy or detect a voice trigger phrase.
Dependencies are inferred from the claim text.