A review of available audio-visual speech corpora and a description of a new multimodal corpus of English speech recordings is provided. The new corpus containing 31 hours of recordings was created specifically to assist audio-visual speech recognition systems (AVSR) development. The database related to the corpus includes high-resolution, high-framerate stereoscopic video streams from RGB cameras, depth imaging stream utilizing Time-of-Flight camera accompanied by audio recorded using both: a microphone array and a microphone built in a mobile computer. For the purpose of applications related to AVSR systems training, every utterance was manually labeled, resulting in label files added to the corpus repository. Owing to the inclusion of recordings made in noisy conditions the elaborated corpus can also be used for testing robustness of speech recognition systems in the presence of acoustic background noise. The process of building the corpus, including the recording, labeling and post-processing phases is described in the paper. Results achieved with the developed audio-visual automatic speech recognition (ASR) engine trained and tested with the material contained in the corpus are presented and discussed together with comparative test results employing a state-of-the-art/commercial ASR engine. In order to demonstrate the practical use of the corpus it is made available for the public use.

Evaluation of sound event detection, classification and localization of hazardous acoustic events in the presence of background noise of different types and changing intensities is presented. The methods for discerning between the events being in focus and the acoustic background are introduced. The classifier, based on a Support Vector Machine algorithm, is described. The set of features and samples used for the training of the classifier are introduced. The sound source localization algorithm based on the analysis of multichannel signals from the Acoustic Vector Sensor is presented. The methods are evaluated in an experiment conducted in the anechoic chamber, in which the representative events are played together with noise of differing intensity. The results of detection, classification and localization accuracy with respect to the Signal to Noise Ratio are discussed. The results show that the recognition and localization accuracy are strongly dependent on the acoustic conditions. We also found that the engineered algorithms provide a sufficient robustness in moderately intense noise in order to be applied to practical audio-visual surveillance systems.

In this article, the problem of creating a safe pedestrian detection model that can operate in the real world is tackled. While recent advances have led to significantly improved detection accuracy on various benchmarks, existing deep learning models are vulnerable to invisible to the human eye changes in the input image which raises concerns about its safety. A popular and simple technique for improving robustness is using data augmentation. In this work, the robustness of existing data augmentation techniques is evaluated to propose a new simple augmentation scheme where during training, an image is combined with a patch of a stylized version of that image. Evaluation of pedestrian detection models robustness and uncertainty calibration under naturally occurring corruption and in realistic crossdataset evaluation setting is conducted to show that our proposed solution improves upon previous work. In this paper, the importance of testing the robustness of recognition models is emphasized and it shows a simple way to improve it, which is a step towards creating robust pedestrian and object detection models.

A method for automatic determination of position of chosen sound events such as speech signals and impulse sounds in 3-dimensional space is presented. The events are localized in the presence of sound reflections employing acoustic vector sensors. Human voice and impulsive sounds are detected using adaptive detectors based on modified peak-valley difference (PVD) parameter and sound pressure level. Localization based on signals from the multichannel acoustic vector probe is performed upon the detection. The described algorithms can be employed in surveillance systems to monitor behavior of public events participants. The results can be used to detect sound source position in real time or to calculate the spatial distribution of sound energy in the environment. Moreover, the spatial filtration can be performed to separate sounds arriving from a chosen direction.

Artificial Neural Network (ANN) models, specifically Convolutional Neural Networks (CNN), were applied to extract emotions based on spectrograms and mel-spectrograms. This study uses spectrograms and mel-spectrograms to investigate which feature extraction method better represents emotions and how big the differences in efficiency are in this context. The conducted studies demonstrated that mel-spectrograms are a better-suited data type for training CNN-based speech emotion recognition (SER). The research experiments employed five popular datasets: Crowd-sourced Emotional Multimodal Actors Dataset (CREMA-D), Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), Surrey Audio-Visual Expressed Emotion (SAVEE), Toronto Emotional Speech Set (TESS), and The Interactive Emotional Dyadic Motion Capture (IEMOCAP). Six different classes of emotions were used: happiness, anger, sadness, fear, disgust, and neutral. However, some experiments were prepared to recognize just four emotions due to the characteristics of the IEMOCAP dataset. A comparison of classification efficiency on different datasets and an attempt to develop a universal model trained using all datasets were also performed. This approach brought an accuracy of 55.89% when recognizing four emotions. The most accurate model for six emotion recognition was trained and achieved 57.42% accuracy on a combination of four datasets (CREMA-D, RAVDESS, SAVEE, TESS). What is more, another study was developed that demonstrated that improper data division for training and test sets significantly influences the test accuracy of CNNs. Therefore, the problem of inappropriate data division between the training and test sets, which affected the results of studies known from the literature, was addressed extensively. The performed experiments employed the popular ResNet18 architecture to demonstrate the reliability of the research results and to show that these problems are not unique to the custom CNN architecture proposed in experiments. Subsequently, the label correctness of the CREMA-D dataset was studied through the employment of a prepared questionnaire.