Assessment of lung sounds (LS) and documentation of their characteristics is a part of routine pulmonary diagnostic procedures. Peculiar acoustic components associated with inspiratory and/or expiratory phases of respiration such as for example asthmatic wheezes serve as useful diagnostic indicators. Supplementing manual auscultation with automatic detection and categorization methods provides opportunities of enhanced long term monitoring which is indispensable in COPD or other respiratory diseases. Also, due to age related or other sensorial limitations of a health care provider, automatic event detection of LS could significantly enhance these aspects of medical diagnosis. In particular, proposed novel method is based on the elements of VAD (Voice Activity Detection), associated with similar to LS frequency components, and utilization of Gaussian Mixture Models (GMM). The automatic detection with GMM is supported by Mel Frequency Cepstral Coefficients (MFCC). This technique allows detecting and classifying peculiar inhalation and exhalation related events of LS, and obtain a higher classification rates with other VAD forms. The efficiency classification achieved with parallel Hidden Markov Models (HMM) was 95% documenting its viable clinical suitability.