Machine learning based sound field analysis
AssigneeApple Inc.
InventorsJoshua D. Atkins, Mehrez Souden, Symeon Delikaris-Manias, Peter Raffensperger
Life of this patent
Estimating the term…
Abstract
Impulse responses of a device are measured. A database of sound files is generated by convolving source signals with the impulse responses of the device. The sound files from the database are transformed into time-frequency domain. One or more sub-band directional features is estimated at each sub-band of the time-frequency domain. A deep neural network (DNN) is trained for each sub-band based on the estimated one or more sub-band directional features and a target directional feature.
Claims27
A method for performing machine based learning for sound field analysis for a device, the method comprising:
- measuring impulse responses of the device;
- generating a database of sound files by convolving source signals with the impulse responses of the device to produce the sound files;
- transforming the sound files from the database into time-frequency domain;
- estimating one or more sub-band directional features at each sub-band of the transformed sound files from the database in the time-frequency domain;
- and
- training a deep neural network (DNN) for each sub-band based on the estimated one or more sub-band directional features and a target directional feature.
An audio system comprising:
- a processor;
- memory having stored therein instructions for performing machine based learning for sound field analysis for a device, the instructions when executed by the processor cause the processor to:
- measure impulse responses of the device;
- generate a database of sound files by convolving source signals with the impulse responses of the device to produce the sound files;
- transform the sound files from the database into time-frequency domain;
- estimate one or more sub-band directional features at each sub-band of the transformed sound files from the database in the time-frequency domain;
- and
- train a deep neural network (DNN) for each sub-band based on the estimated one or more sub-band directional features and a target directional feature.
A non-transitory computer-readable storage medium storing executable program instructions for performing machine based learning for sound field analysis for a device, the program instructions when executed by a processor cause the processor to perform a method comprising:
- measuring impulse responses of the device;
- generating a database of sound files by convolving source signals with the impulse responses of the device to produce the sound files;
- transforming the sound files from the database into time-frequency domain;
- estimating one or more sub-band directional features at each sub-band of the transformed sound files from the database in the time-frequency domain;
- and
- training a deep neural network (DNN) for each sub-band based on the estimated one or more sub-band directional features and a target directional feature.
Dependencies are inferred from the claim text.