Wyniki wyszukiwania - BazTech

1

Machine learning-based analysis of English lateral allophones

Piotrowska Magdalena, Korvel Gražina, Kostek Bożena, Ciszewski Tomasz, Czyżewski Andrzej

International Journal of Applied Mathematics and Computer Science

|

2019

|

Vol. 29, no. 2

393--405

EN

Automatic classification methods, such as artificial neural networks (ANNs), the k-nearest neighbor (kNN) and self-organizing maps (SOMs), are applied to allophone analysis based on recorded speech. A list of 650 words was created for that purpose, containing positionally and/or contextually conditioned allophones. For each word, a group of 16 native and non-native speakers were audio-video recorded, from which seven native speakers’ and phonology experts’ speech was selected for analyses. For the purpose of the present study, a sub-list of 103 words containing the English alveolar lateral phoneme /l/ was compiled. The list includes ‘dark’ (velarized) allophonic realizations (which occur before a consonant or at the end of the word before silence) and 52 ‘clear’ allophonic realizations (which occur before a vowel), as well as voicing variants. The recorded signals were segmented into allophones and parametrized using a set of descriptors, originating from the MPEG 7 standard, plus dedicated time-based parameters as well as modified MFCC features proposed by the authors. Classification methods such as ANNs, the kNN and the SOM were employed to automatically detect the two types of allophones. Various sets of features were tested to achieve the best performance of the automatic methods. In the final experiment, a selected set of features was used for automatic evaluation of the pronunciation of dark /l/ by non-native speakers.

2

Machine learning-based novel approach to classify the shoulder motion of upper limb amputees

Amanpreet Kaur

Biocybernetics and Biomedical Engineering

|

2019

|

Vol. 39, no. 3

857--867

EN

An upper limb amputation is a traumatic event that can seriously affect the person's capacity to perform regular tasks and can lead individuals to lose their confidence and autonomy. Prosthetic devices can be controlled via the acquisition and processing of electromyogram signal produced at the muscles fiber from the surface of the body with an array of an electrode placed on the residual limb. This paper presents the feasibility of classifying the different shoulder movements from around shoulder muscles of transhumeral amputees. The performance of a classifier is affected by the variation of Surface Electromyography (sEMG) signals due to the different categories of contraction. To avoid this, the wavelet transform and data transformation method are employed for features extraction from sEMG signals. Afterward five different supervised machine learning techniques viz. Support Vector Machines (SVM), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbor (KNN) and Naïve Bayes (NB) are applied to determine the different classifiers accuracy. An effective combination of wavelet and RF achieves the best performance with a total classification accuracy of 98%.

3

Pattern layer reduction for a generalized regression neural network by using a self-organizing map

Kartal S., Oral M., Ozyildirim B. M.

International Journal of Applied Mathematics and Computer Science

|

2018

|

Vol. 28, no. 2

411--424

EN

In a general regression neural network (GRNN), the number of neurons in the pattern layer is proportional to the number of training samples in the dataset. The use of a GRNN in applications that have relatively large datasets becomes troublesome due to the architecture and speed required. The great number of neurons in the pattern layer requires a substantial increase in memory usage and causes a substantial decrease in calculation speed. Therefore, there is a strong need for pattern layer size reduction. In this study, a self-organizing map (SOM) structure is introduced as a pre-processor for the GRNN. First, an SOM is generated for the training dataset. Second, each training record is labelled with the most similar map unit. Lastly, when a new test record is applied to the network, the most similar map units are detected, and the training data that have the same labels as the detected units are fed into the network instead of the entire training dataset. This scheme enables a considerable reduction in the pattern layer size. The proposed hybrid model was evaluated by using fifteen benchmark test functions and eight different UCI datasets. According to the simulation results, the proposed model significantly simplifies the GRNN’s structure without any performance loss.