As an additional example, computing device 106 can receive the audio sensed by audio sensors 104-1, 104-2, . . . , 104-M during an event associated with (e.g., that occurs during) the operation of HVAC equipment 102-1, 102-2, . . . , 102-N, and send, via network 108, this audio to computing device 110. Computing device 110 can provide this audio to a user (e.g., technician) to determine whether the event corresponds to a fault occurring in equipment 102-1, 102-2, . . . , 102-N. For example, computing device 110 can provide an alert of the event to the user and play the audio sensed during the event for the user, who can listen to it and determine (e.g., based on the user's expert knowledge) whether it corresponds to a fault occurring in the equipment.

Upon the user (e.g., technician) determining the event corresponds to a fault occurring in the equipment, computing device 110 can receive from the user an input indicating that the audio sensed during the event corresponds to the fault. For instance, the user may identify (e.g., annotate) the audio sensed during the event as corresponding to the fault, and store this identified audio (e.g., the profile of the audio) accordingly.

As such, computing device 110 can learn audio that corresponds to a fault (e.g., a particular type of fault) in a user (e.g., technician) supervised manner. That is, computing device 110 can capture the technician's expert knowledge in determining whether noise in the audible range corresponds to a fault. For example, computing device 110 can use supervised machine learning to classify audio as faults, and build an audio corpus (e.g., database of audio files) for different classes of faults.