Three main approaches on how audio signals can be used as input to a deep learning model are: extracting hand-crafted features from audio signals, mapping audio signals into appropriate images such as spectrogram-like ones, and using directly raw audio signals. Among these approaches, the usage of spectrogram-like images represents a compromise regarding the bias enforced by the processing (seen in hand-crafted features) and computational demands (seen in raw audio signals). When any of the spectrogram-like images is used as a deep learning model input, then different techniques for image processing become available and can be implemented. They include techniques for assessing the image similarity, implementing image matching, and image recognition. The topic of this paper is similarity of spectrogram-like images obtained from DC motor sounds. In that respect, relevant measures of image similarity are first reviewed, and then one of them - the Pearson correlation coefficient - is applied for evaluating the similarity within the same class and between two classes of different spectrogram-like images.

Noise represents one of the most significant disturbances in measured room impulse responses (RIRs), and it has a potentially large impact on evaluation of the decay parameters. In order to reduce noise effects, various methods have been applied, including truncation of an RIR. In this paper, a procedure for the response truncation based on a model of RIR (nonlinear decay model) is presented. The model is represented by an exponential decay plus stationary noise. Unknown parameters of the model are calculated by an optimization that minimizes the difference between the curve generated by the model and the target one of the response to be truncated. Different curves can be applied in the optimization—absolute value of the RIR, logarithmic decay curve, and Schroeder curve obtained by the backward integration of the RIR. The proposed procedure is tested on various synthesized and measured impulse responses. It is compared with the procedure taken from the literature, often applied in practice.

The Zero Crossing Signature (ZCS) approach is a novel time-domain feature extraction method that analyzes zero crossing points over multiple amplitude-shifted versions of an acoustic signal, enabling richer information extraction while maintaining computational efficiency. This method is especially suitable for real-time classification in the emerging Internet of Things (IOT) landscape, where resource-constrained devices require low-power solutions to support emission reduction efforts. In this study, the ZCS method was employed to showcase its full potential by classifying vehicles as diesel or gasoline based on their acoustic signatures. This classification task, applied to a database of car sounds acquired in the authors’ previous research, serves as a comprehensive demonstration of the method’s capabilities in distinguishing between engine types through characteristic sound wave patterns, highlighting its effectiveness and applicability in real-world scenarios. To further enhance feature extraction while keeping computational costs low, simple transformations using the first and second derivatives of the acoustic signals were applied, offering an efficient means of capturing additional signal characteristics. A dataset of 417 vehicle recordings was analyzed, and the classification performance of ZCS was compared with the conventional Zero Crossing (ZC) method using a Self-Organizing Map (SOM) configured with a 1D grid of 9 neurons. The study evaluated various time constants and crossing threshold densities for ZCS, benchmarking them against the classical ZC approach to assess their effectiveness.

In this paper a portable system for acquisition of sound generated by passenger vehicles powered by internal combustion engines is described and analyzed. The acquisition system is developed from scratch and tested in order to satisfy the requirements such as high-quality of audio recordings, high mobility, robustness and privacy respect. With this acquisition system and adequate signal processing, the main goal was to collect a large amount of clear audio recordings that will form a quality dataset. In further research, this dataset will be used for machine learning model training and testing, i.e. for developing a system for automatic recognition of the type of car engine based on fuel.