Streaming media applications generate a sizable part of network traffic and represent a significant proportion of network providers' income. The commitment to user satisfaction can be summarized by different concepts, content providers emphasizing quality of experience (QoE), whereas network providers are more focused on quality of service (QoS). Measuring QoS parameters, and understanding the relationship between the two, is essential to enable network tuning for enhanced QoE. Analysis of actual streaming dynamics and detection of impairments require the ability to discriminate between audio and video packet flows. For the purpose we present a cross-layer analysis based on the study of packet flow features obtained by implementing a Support Vector Machine. This paper presents results of a study supported by the use of nTh HighSee, a non-intrusive QoS monitoring tool, where approaches to video/audio detection have been investigated and tested.