Disney patents audio AI to detect coughs and machine faults

1. A system comprising: a computing platform including a processing hardware and a system memory; a software code including a trained machine learning (ML) model stored in the system memory, the trained ML model including: (i) an input stage created based on a first ML model trained to detect one or more predetermined data attributes, the input stage trained using sliding window convolutions with stride length being a tunable hyperparameter of the input stage and a training dataset generated by a second ML model trained using an output of the first ML model and that has been annotated to identify the one or more predetermined data attributes, to predict bounding timestamps and to detect temporal segments of a test dataset that include the predetermined one or more data attributes, and (ii) an output stage trained using another training dataset that has been annotated to correlate the one or more predetermined data attributes with one of a condition or a property; the processing hardware configured to execute the software code to: receive a dataset; perform an analysis of the dataset, using the input stage of the trained ML model, to detect a presence of at least one predetermined data attribute, the at least one predetermined data attribute comprising an audio attribute including a voiced bilabial nasal sound or audio content corresponding to operation of a machine, wherein performing the analysis comprises automatically segmenting the dataset by identifying one or more temporal segments that include the at least one predetermined data attribute and extracting the one or more temporal segments based on bounding timestamps predicted by the input stage; and: (a) predict, by applying the output stage to the extracted one or more temporal segments when the analysis of the dataset detects the audio attribute including the voiced bilabial nasal sound, a probability that the at least one predetermined data attribute is indicative of a respiratory condition or an infectious disease status of a human subject, or (b) predict, by applying the output stage to the extracted one or more temporal segments when the analysis of the dataset detects the audio attribute including the audio content corresponding to the operation of the machine, a probability that the at least one predetermined data attribute is indicative of a malfunction by the machine.