RHRobert Hossa

Robert Hossa

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Robert Hossa has a total of 5 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • The Impact of Training Strategies on Overfitting in Vowel Classification Using PS-HFCC Parametrization for Automatic Speech Recognition

    by Stanislaw Gmyrek, Urszula Libal, Robert Hossa

    This paper investigates the overfitting problem in vowel classification task for automatic speech recognition (ASR). It utilizes a pitch synchronized human factor cepstral coefficients (PS-HFCC) as the parametrization method, which outperforms traditional methods like HFCC and mel-frequency cepstral coefficients (MFCC) in frame-level classification accuracy. While deep learning models are prevalent in contemporary ASR systems, they often lack explainability, a characteristic of classical classifiers. Therefore, this study examines overfitting phenomenon using a range of classifiers with well-understood properties. Specifically, it analyzes the impact of different training strategies on classifier performance, comparing the susceptibility to overfitting of several widely used classifiers, including the Gaussian mixture model (GMM), a standard approach in speech recognition. The analysis of training strategies considers various data splitting methods: random, speaker-based, and cluster-based. Our analysis of training strategies highlights the crucial role of data splitting methods: while random splitting is commonly used, it can lead to inflated accuracy due to overfitting. We demonstrate that speaker-independent splitting, where the classifier is trained on one set of speakers and tested on a separate, unseen set, is essential for robust evaluation and for accurately assessing generalization to new speakers. Potentially, the resulting insights may inform the future development and training of more reliable ASR systems.


  • Robust speech parametrization based on pitch synchronized cepstral solutions

    by Stanisław Gmyrek, Robert Hossa

    In general, the speech signal can be described by the excitation signal, the impulse response of the vocal tract, and a system that describes the impact of speech emission through human lips. The characteristics of the vocal tract primarily shape the semantic content of speech. Regrettably, the irregular periodicity of glottal excitation represents a significant factor in generating substantial distortions (ripples) in the amplitude spectrum of voiced speech. In this study, a PS-STFT (Pitch- Synchronized Short-Time Fourier Transform) method was proposed to achieve a reliable amplitude spectrum of the vocal tract. Subsequently, a set of cepstral coefficient vectors, namely PSHFCC (Pitch Synchronized Human Factor Cepstral Coefficients), as a chosen representative of the commonly used classical cepstral parameterization methods was analyzed to investigate the statistical properties after correction. Additionally, the widely accepted in speech recognition applications, the GMM (Gaussian Mixture Model) was chosen as the statistical acoustic model of individual Polish speech phonemes. To evaluate the quality of the proposed method, the distances between the multivariate probability distributions of the GMM form were calculated. Modifying classical cepstral methods through the analysis of variable-length signal frames synchronized to the fundamental period resulted in a reduction in the variance of the estimators of the cepstral coefficients, leading to an increase in the distances between the probability distributions and, consequently, improved classification results.

  • Open Access

    The Influence of the Amplitude Spectrum Correction in the HFCC Parametrization on the Quality of Speech Signal Frame Classification

    by Stanislaw Gmyrek, Robert Hossa, Ryszard Makowski

    The voiced parts of the speech signal are shaped by glottal pulse excitation, the vocal tract, and the speaker’s lips. Semantic information contained in speech is shaped mainly by the vocal tract. Unfortunately, the quasiperiodicity of the glottal excitation, in the case of the HFCC parameterization, is one of the factors affecting the significant scatter of the feature vector values by introducing ripples into the amplitude spectrum. This paper proposes a method to reduce the effect of quasiperiodicity of the excitation on the feature vector. For this purpose, blind deconvolution was used to determine the vocal tract transfer function estimator and the corrective function of the amplitude spectrum. Subsequently, on the basis of the obtained HFCC parameters, statistical models of individual Polish speech phonemes were developed in the form of mixtures of Gaussian distributions, and the influence of the correction on the quality of classification of speech frames containing Polish vowels was considered in details. The aim of the introduced solution was to narrow the GMM distributions, which clearly, according to the detection theory, reduces classification errors. The results obtained confirm the effectiveness of the proposed method.

  • Open Access

    Amplitude spectrum correction to improve speech signal classification quality

    by Stanislaw Gmyrek, Robert Hossa, Ryszard Makowski

    The speech signal can be described by three key elements: the excitation signal, the impulse response of the vocal tract, and a system that represents the impact of speech production through human lips. The primary carrier of semantic content in speech is primarily influenced by the characteristics of the vocal tract. Nonetheless, when it comes to parameterization coefficients, the irregular periodicity of the glottal excitation is a significant factor that leads to notable variations in the values of the feature vectors, resulting in disruptions in the amplitude spectrum with the appearance of ripples. In this study, a method is suggested to mitigate this phenomenon. To achieve this goal, inverse filtering was used to estimate the excitation and transfer functions of the vocal tract. Subsequently, using the derived parameterisation coefficients, statistical models for individual Polish phonemes were established as mixtures of Gaussian distributions. The impact of these corrections on the classification accuracy of Polish vowels was then investigated. The proposed modification of the parameterisation method fulfils the expectations, the scatter of feature vector values was reduced.

  • Reducing the Impact of Fundamental Frequency on the HFCC Parameters of the Speech Signal

    by Stanislaw Gmyrek, Robert Hossa, Ryszard Makowski

    The voiced parts of the speech signal are shaped by glottal pulse excitation, the vocal tract, and the speaker's lips. Semantic information contained in speech is shaped mainly by the vocal tract. Unfortunately, the quasiperiodicity of the glottal excitation, in the case of Human Factor Cepstral Coefficients (HFCC) parameterization, is one of the factors affecting the significant scatter of the feature vector values by introducing ripples into the amplitude spectrum. This paper proposes a method to reduce the effect of the quasiperiodicity of the excitation on the feature vector. For this purpose, blind deconvolution was used to determine the vocal tract transfer function estimator and the corrective function of the amplitude spectrum. Then, based on the obtained HFCC parameters, statistical models of individual Polish speech phonemes were developed in the form of mixtures of Gaussian distributions, and the influence of the correction on the quality of classification of speech frames containing Polish vowels was investigated.

  • Open Access

    An Effective Speaker Clustering Method using UBM and Ultra-Short Training Utterances

    by Robert Hossa, Ryszard Makowski

    The same speech sounds (phones) produced by different speakers can sometimes exhibit significant differences. Therefore, it is essential to use algorithms compensating these differences in ASR systems. Speaker clustering is an attractive solution to the compensation problem, as it does not require long utterances or high computational effort at the recognition stage. The report proposes a clustering method based solely on adaptation of UBM model weights. This solution has turned out to be effective even when using a very short utterance. The obtained improvement of frame recognition quality measured by means of frame error rate is over 5%. It is noteworthy that this improvement concerns all vowels, even though the clustering discussed in this report was based only on the phoneme a. This indicates a strong correlation between the articulation of different vowels, which is probably related to the size of the vocal tract.


  • Open Access

    Automatic speech signal segmentation based on the innovation adaptive filter

    by Ryszard Makowski, Robert Hossa

    Speech segmentation is an essential stage in designing automatic speech recognition systems and one can find several algorithms proposed in the literature. It is a difficult problem, as speech is immensely variable. The aim of the authors’ studies was to design an algorithm that could be employed at the stage of automatic speech recognition. This would make it possible to avoid some problems related to speech signal parametrization. Posing the problem in such a way requires the algorithm to be capable of working in real time. The only such algorithm was proposed by Tyagi et al., (2006), and it is a modified version of Brandt’s algorithm. The article presents a new algorithm for unsupervised automatic speech signal segmentation. It performs segmentation without access to information about the phonetic content of the utterances, relying exclusively on second-order statistics of a speech signal. The starting point for the proposed method is time-varying Schur coefficients of an innovation adaptive filter. The Schur algorithm is known to be fast, precise, stable and capable of rapidly tracking changes in second order signal statistics. A transfer from one phoneme to another in the speech signal always indicates a change in signal statistics caused by vocal track changes. In order to allow for the properties of human hearing, detection of inter-phoneme boundaries is performed based on statistics defined on the mel spectrum determined from the reflection coefficients. The paper presents the structure of the algorithm, defines its properties, lists parameter values, describes detection efficiency results, and compares them with those for another algorithm. The obtained segmentation results, are satisfactory.


  • Holographic antenna diagnosis with a bi-polar near-field measuring system

    by R. Hossa, P. Kabacik

    Research on advanced antenna array designs calls for sophisticated measuring techniques to diagnose the actual properties of the array models. These methods should provide not only the characteristics of the AUT (expressed in terms of standard parameters), but also the actual current distributions for the array aperture. Knowing the amplitude and phase distributions of the currents in the antenna aperture, we can investigate the troublesome issues deteriorating the quality of the antenna properties. We present some major results of our research which aimed at the developing diagnostic tools suitable for implementation in a bi-polar near-field facility.