Systems and methods are configured to process audio signals to identify content-types. Audio content is received at an audio decoder which decodes the audio content. The decoded audio content is segmented into frames by applying a windowing function to a given audio frame using a window having a time width related to a delay time of the decoder. A power spectrum estimate of a given frame is determined. A mel filter bank is applied to the power spectrum of the frame. A DCT matrix is applied to filter bank energies to generate a DCT output. A log of the DCT output is used to generate a mel coefficient 1. A threshold for the content is dynamically determined. The mel coefficient 1 and the dynamically determined threshold are used to detect a near silence between content-types and to identify the content-types.