Narrow-band autocorrelation function features for the automatic recognition of acoustic environments

Abstract

Acoustic environments are typically composed of multiple sound sources of different typologies, making them especially complex to model and parameterize. To develop an automatic acoustic environment recognition system, this work proposes a spectro-temporal signal parameterization technique inspired by human perception. The proposed parameters are derived from the analysis of the autocorrelation function of narrow-band signals (NB-ACF) obtained from an auditory gammatone filter bank. Five features related to acoustic phenomena are extracted from the NB-ACF to parameterize sound signals. Experiments are conducted on a 4 h sound database (composed of 15 different acoustic environments) employing four different machine learning techniques: K-nearest neighbor, Gaussian mixture models, neural networks and support vector machines. The averaged recognition rates increase by 4.5% when using the proposed NB-ACF features instead of the popular Mel frequency cepstral coefficients. The improvement is even greater (5.6%) when the features are tested in acoustic environments from unknown locations. These results are attributed to the better modeling of the acoustic environments thanks to the complementarity of the spectro-temporal features derived from NB-ACF analysis.

Data Augmentation and Deep Learning Methods in Sound Classification: A Systematic Review

“…Class-dependent temporal-spectral structures and long-term descriptive statistics features were extracted for sound events. Other authors applied the Discrete Gabor Transform (DGT) audio image representation [119], multiresolution feature [53], hybrid method based on mel frequency cepstral coefficient and the gammatone frequency cepstral coefficient [62], inverted MFCC and extended MFCC [66], bag of audio words (BoAW) [120], narrow band auto-correlation features (NB-ACF) [121].…”

A Review of Physical and Perceptual Feature Extraction Techniques for Speech, Music and Environmental Sounds

“…These features have been shown to provide good performance for indoor and outdoor environmental sound classification. In [161], the same authors improved this technique by substituting the Mel filter bank employed to obtain the narrow-band signals by a Gammatone filter bank with Equivalent Rectangular Bandwidth bands. In addition, the Autocorrelation Zero Crossing Rate (AZCR) was added, following previous works like the one by Ghaemmaghami et al [43].…”

“…• Kernel Power Flow Orientation Coefficients (KPFOC): in the works by Gerazov and Ivanovski [184,185], a bank of 2D kernels is used to estimate the orientation of the power flow at every point in the auditory spectrogram calculated using a Gammatone filter bank (Valero and Alías [161]), obtaining an ASR front-end with increased robustness to both noise and room reverberation with respect to previous approaches, and specially for small vocabulary tasks.…”

Environmental sound recognition: a survey

“…On the other hand, the auto-correlation function (ACF) represents the time-evolution and has an intimate relationship with the power spectral density (PSD) of the underlying signal. Valero and Alias [26] proposed a new set of features called the Narrow-Band Auto Correlation Function features (NB-ACF). The extraction of NB-ACF features can be explained using Fig.…”

“…This demands a sub-frame length to be much larger than that used in sub-framing processing. It is recommended in [26] that a be sub-frame of size 500 ms with an overlap of 400 ms in a frame of 4 s. Finally, KNN and SVM classifiers are used for decision making in each sub-frame. The performance of NB-ACF features was compared with MFCC and discrete wavelet transform (DWT) coefficients with a data-set consisting of 15 environmental scenes.…”