|
Exploring the relationship between the usability in use and the adoption of Google Classroom Paul Stefan Popescu, Costin Pribeanu |
23-37 |
|
Enhancing an End-to-end Romanian Speech Recognition Model for Persons with Speech Impairments Elena Pelican, Lucian Odainic |
38-50 |
1 University of Craiova, Department of Computers and Information Technology
Blvd. Decebal nr. 107, RO-200440, Craiova, Romania
2 Academy of Romanian Scientists
Splaiul Indpendentei 54, Bucharest, Romania
Abstract. People’s lives changed during the pandemic. Because of the lockdown restrictions universities had to shift from face-to-face teaching and learning to distance education. Since online platforms became the working space for educational activities usability and quality in use became critical issues for technology adoption. However, the relationship between technology adoption, usability, and quality in use has been rarely investigated in extant research. The objective of this research is to develop and test a model that measures the influence of usability in use on the intention to use the Google Classroom platform after the pandemic. The model includes five determinants of continuance intention: extrinsic motivation, intrinsic motivation, effectiveness in use, efficacy in use, and satisfaction. The results show that the model explains a lot of variances in satisfaction and continuance intention. Intrinsic motivation and effectiveness in use were the most important factors.
Keywords: Quality in use, usability in use, TAM, Google Classroom, COVID-19.
Cite this paper as:
Popescu, P. S., Pribeanu, C. Exploring the relationship between the usability in use and the adoption of Google Classroom.
International Journal of User-System Interaction 15(2),
23-37, 2022.
Ovidius University Constanta
124 Mamaia Blvd., Constanta, Romania
Abstract. In this paper we propose an approach to improve a deep learning model, built for speech recognition in Romanian language. The improvements have been made for serving a person with speech difficulties, as already existing model have proved unsatisfactory results. We have updated the weights on new collected data from the person in question. In order to make the model more robust, audio data was modified by adding noise and changing the speed by making it slower and then faster. The system relies on a deep learning model similar to DeepSpeech2 that contains convolutional and recurrent neural networks. The layer with CNN requires image data, so the raw recorded waveform was converted into Mel spectrograms, a visual representation of the sound. The resulting model has shown to be more appropriate for solving the problem under consideration, so we have incorporated it within an Android application. This generates transcriptions of audio files that are registered in real time containing Romanian speech. The Android application was inspired by Google Live Transcribe, which, when testing it, the transcriptions had imperfections. We thought of colliding the enhanced model and the idea of a live transcriber to have the solution that outperforms the existing products and suits perfectly our case.
Keywords: Computational linguistics, Deep learning, Romanian language, Speech recognition, Speech-to-text
Cite this paper as:
Pelican, E., Odainic, L. Enhancing an End-to-end Romanian Speech Recognition Model for Persons with Speech Impairments.
International Journal of User-System Interaction 15(2),
38-50, 2022.