In the conduct of this project, the proponents primarily aimed to explore the threshold for the deviations of the Filipino-accented utterances of selected English words using the MFCC and DTW concepts. The initial premise utilized by the proponents would be the speaker-dependent nature of the MFCC; hence, the calculations, measurements, and data-gathering methodologies were conducted by means of acquiring the said coefficients from the same individual verbally uttering selected words in that of the American accent and in their native Filipino accent and subjecting these results to a series of MA TLAB algorithms devised by the researchers. As such, the study was able to conclude that, upon preliminary calculations, the normalized DTW threshold between the Filipino-Accented English was calculated to be 4.91 with the designed system having an accuracy of 68.73 % in correctly determining which Filipino-accented utterances correspond to their respective English word counterparts. While this was able to procure plausible results, one of the limitations observed in this implementation would be the presence of noise in the samples that may have caused deviations along with the limited number of participants that partook in the acquisition of data for this study. Thus, it is then highly suggested that a wider and more robust database be implemented in future studies involving this subject and relative methodologies.
Baybayin is an old Philippine script that has become part of the Filipino heritage; however, due to the difficulty of the writing system, most people nowadays do not use it anymore. That is why this paper proposed an improved CNN model for classifying the Philippine Script Baybayin in order to help people to learn this old script. The model was systematically developed through experimentation and hyperparameter tuning with MobileNetV2 as the baseline. Sample handwritten Baybayin characters were obtained from Kaggle and Mendeley Data for training. It was evaluated that the proposed model has training and validation accuracies of 97.04% and 96.02%, respectively, in contrast to the MobileNetV2, which has slightly lower accuracies and larger fluctuations in its validation accuracy. Aside from that, this model only has total parameters of 892,255 and a size of 3575.944 kB. Lastly, through mobile deployment testing, it was concluded that only the proposed model is working, as the MobileNetV2 was not able to accurately detect the diacritics.
Graph neural networks (GNNs) have gained significant attention in recent years for their ability to process data that may be represented as graphs. This has prompted several studies to explore their representational capability based on the graph isomorphism task. These works inherently assume a countable node feature representation, potentially limiting their applicability. Interestingly, only a few study GNNs with uncountable node feature representation. In the paper, a novel perspective on the representational capability of GNNs is investigated across all levels—node-level, neighborhood-level, and graph-level—when the space of node feature representation is uncountable. More specifically, the strict injective and metric requirements are softly relaxed by employing a pseudometric distance on the space of input to create a soft-injective function such that distinct inputs may produce similar outputs if and only if the pseudometric deems the inputs to be sufficiently similar on some representation. As a consequence, a simple and computationally efficient soft-isomorphic relational graph convolution network (SIR-GCN) that emphasizes the contextualized transformation of neighborhood feature representations via anisotropic and dynamic message functions is proposed. A mathematical discussion on the relationship between SIR-GCN and widely used GNNs is then laid out to put the contribution into context, establishing SIR-GCN as a generalization of classical GNN methodologies. Experiments on synthetic and benchmark datasets then demonstrate the relative superiority of SIR-GCN, outperforming comparable models in node and graph property prediction tasks.
