Classifying User Environment for Mobile Applications using Linear Autoencoding of Ambient Audio

Abstract

Many mobile devices and applications can act in contextsensitive ways, but rely on explicit human action for context awareness. It would be preferable if our devices were able to attain context awareness without human intervention. One important aspect of user context is environment. We present a novel method for classifying environment types based on acoustic signals. This method makes use of linear autoencoding neural networks, and is motivated by the observation that biological coding systems seem to be heavily influenced by the statistics of their environments. We show that the autoencoder method achieved a lower error rate than a standard gaussian mixture model on a representative sample task, and that a linear combination of autoencoders and GMMs yielded better performance than either alone.

Environment Recognition for Digital Audio Forensics Using MPEG-7 and MEL Cepstral Features

“…The autoencoder and GMM achieved 77.9 % and 77.57 % accuracy, respectively, in the experiments reported in [10], while a hybrid system between them provided 80.05 % accuracy. MFCC and 11-state HMMs gave 91.5 % average accuracy for 14 classes in [11].…”

“…Malkin and Waibel [10] introduced linear autoencoding neural networks for classifying environment. The autoencoder is a standard feed-forward neural network with a linear transform function.…”

“…Eronen et al [6] used PCA (principal component analysis), ICA (independent component analysis), and LDA (linear discriminated analysis), while Malkin and Waibel [10], and Ntalampiras et al [13] applied PCA. Zeng [7] used greedy method to reduce the number of feature dimension.…”

“…It was claimed that features with Gabor properties could provide a flexible representation of time and frequency localization of unstructured sounds like environment. MFCCs and spectral centroid were used as features in [10], while [11] used MFCC features only. Unlike [6, 8, 11, 7, 12, and 13], they used forensics application like audio files, where both ambient (environmental) sound and human speech were present.…”

Audio Environment Recognition using Zero Crossing Features and MPEG-7 Descriptors

“…Sixty-four dimensional MFCC, plus the spectral centroid were used as features in (Malkin and Waibel, 2005). They used forensic-application-like audio files, where both ambient, i.e., environmental sound and human speech were present.…”