CPC H04R 3/00 (2013.01) [G06N 3/08 (2013.01); H04R 2420/07 (2013.01)] | 20 Claims |
1. A repetitive noise identification method comprising the steps of:
obtaining a set of ambient sound data;
obtaining a set of sample target noises containing at least one sample target noise;
creating a set of cyclically-shifted target noises by cyclically shifting each of the sample target noises by one or more time-shift intervals;
creating a set of training target noises comprising the set of sample target noises and the set of cyclically-shifted target noises;
training a convoluted neural network to identify one or more target noises using the set of training target noises and the set of ambient sound data;
implementing the convoluted neural network in a monitoring device, the monitoring device having a microphone, a processing unit, a storage unit storing the convoluted neural network, and a wireless communications unit;
deploying the monitoring unit in an area to be monitored for the target sounds;
obtaining an ambient noise sample having an interval time length TA with the microphone;
analyzing the ambient noise sample with the processing unit to determine whether the ambient noise sample includes any of the target noises and if so, sending a data signal containing alert data to a concentrator, otherwise returning to the previous step;
sending the data signal to a cloud-based platform from the monitoring device;
sending the data signal from the cloud-based platform to a backend server;
storing the alert data in a database; and
publishing the alert data to one or more users through one or more access devices.
|