China
Africa
Explore chapters and articles related to this topic
11 Internet Video Technologies
Published in Wes Simpson, Howard Greenfield, IPTV and Internet Video:, 2012
Wes Simpson, Howard Greenfield
Several different file formats are commonly used for Windows Media content, including the following.Windows Advanced Systems Format File (.asf): A file format designed specifically for the transport and storage of content intended for streaming applications.Windows Media Audio File (.wma): A file that contains audio signals encoded using the Windows Media Audio compression system and is in the ASF file format.Windows Media Video File (.wmv): A file that contains video and audio signals encoded using the Windows Media Video and Windows Media Audio compression system and is in the ASF file format.
Detection of AAC compression using MDCT-based features and supervised learning
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2022
José Juan García-Hernández, Wilfrido Gómez-Flores
Regarding AAC compression detection, Luo et al. (2014) proposed using the zero count of the MDCT and Mel-frequency cepstral coefficients for achieving compression history detection. The Support Vector Machine (SVM) classifier performed adequately in the detection of low and medium quality (32 to 128 kbps) coding for AAC as well as MP3 and Windows Media Audio (WMA). An approach to identify fake-quality WAV audio based on phase information was proposed by Zhou et al. (2015). In this method, the differences between the phase of the suspect audio and its compressed version were the input features of an SVM classifier. Experiments for AAC, MP3, and OGG compression detection were presented for distinct quality bitrates (32 to 192 kbps). Seichter et al. (2016) proposed using a convolutional neural network (CNN) for bitrate detection in AAC compression. The procedure consisted of two stages: (1) framing grid detection and (2) zero count of MDCT coefficients, each one contributing to the overall classification error. From the quantised MDCT coefficients, the CNN extracts 3600 features as input to the softmax layer for classifying nine classes of bitrates in the range of 32 to 320 kbps. This method was the first to explore compression detection for high bitrates (i.e., kbps), which is challenging because of the similarity of compressed signals and raw signals increases. Nevertheless, this approach requires accurately determining the window shape, sequence, and grid used during encoding to construct the signal feature. Notice that in scenarios where some suspicious audio samples are lost or only some discontinuous sample blocks are available for forensic analysis that synchronisation requirement is a critical drawback. Recently, Derrien (2019) proposed an AAC compression detection based on quantisation errors in the time-frequency domain without machine learning. This method explores a threshold for rounding error according to distributions for previously compressed and lossless audio. Threshold detection performed well for different bitrates (128 to 320 kbps), although the method lacks double compression detection.