Often, faulty operation is accompanied by electrical discharge processes. Commonly, partial discharges cause noise. This is the key for the use of acoustical methods in maintenance in order to prevent catastrophic failures. Typical discharges are from corona, arcing or tracking type. This paper deals with the improvement of acoustic measurement techniques and equipment for air transmitted acoustic signals. For some reasons, the use of ultrasound is advantageous for the detection and monitoring of partial discharges. Focus of the contribution will be on advanced signal processing strategies for real-time and post-processing of data. In order to develop improved algorithms, a series of experiments has been done under laboratory (in order to understand the physically caused possibilities and limitations for the methodical developments) and field conditions (in order to have access to the variety of the stochastic discharge processes). One of the objectives is the automatic classification of the various types of discharges by means of neural networks without having expert knowledge. It can be complicate to find highly transient signals by a single sensor technique. Discharge signals are highly transient in time and locally fluctuating. Therefore, acoustical imaging is used as a further promising technology.