|
Text generation and music composition by human and neural network interaction Răzvan Păroiu, Ştefan Trăuşan-Matu |
61-72 |
|
Voice-controlled 3D modelling with an intelligent personal assistant Sonia Grigor, Constantin Nandra, Dorian Gorgan |
73-88 |
|
Ear Trainer for Guitar Chords – An Android Application for Recognition of Guitar Chords Matei-Alexandru Cioată, Adrian Iftene |
89-109 |
|
Automatic black and white image colorization using generative adversarial networks Gheorghe-Cosmin Petre, Ştefan Trăuşan-Matu |
110-120 |
1 University Politehnica of Bucharest
313 Splaiul Independentei, Bucharest, Romania
2 Institutul de Cercetări în Inteligenţa Artificială
Calea 13 Septembrie nr. 13, Bucureşti
3 Academy of Romanian Scientists
Splaiul Indpendentei 54, Bucharest, Romania
Abstract. Artificial intelligence has been used for a long time to generate natural language text or music, artifacts that are usually considered as the creation of human minds. However, text and music generated with artificial intelligence in general lack the feeling specific to human creations. Therefore, hybrid approaches where the user generates content by interacting with the artificial intelligence have provided better results. This paper presents an application that follows this hybrid approach, which can be used by a human to create text and compose music with the help of artificial intelligence. Text or musical notes are generated by multiple trained neural network models and the user is the one that selects the best-generated result.
Keywords: artificial creativity, natural language processing, language models, music composition
Cite this paper as:
Păroiu, R., Trăuşan-Matu, S. Text generation and music composition by human and neural network interaction.
International Journal of User-System Interaction 13(2),
61-72, 2020.
Technical University of Cluj-Napoca, Computer Science Department
26-28 G. Baritiu, 400027, Cluj-Napoca, Romania
Abstract. In this paper we look into the feasibility of controlling the process of 3D modelling with the help of voice commands. The main motivation is represented by the potential increase in the accessibility of cluttered graphical interfaces to novice users, helping them to smoothen the learning curve. Throughout this paper, we present a solution that we have implemented to augment the interface of an existing 3D modelling tool for the purpose of creating and editing 3D objects. We describe the overall architecture of the solution, detailing the structure of the command set it employs and the main challenges it was designed to overcome. Finally, we seek to demonstrate the potential of the voice-interaction model by employing its functionality in a 3D modelling use-case and analyzing its relative performance when compared to the use of the graphical interface.
Keywords: voice based user interaction, 3D graphics editing, personal assistant, Amazon Web Services
Cite this paper as:
Grigor, S., Nandra, C., Gorgan, D. Voice-controlled 3D modelling with an intelligent personal assistant.
International Journal of User-System Interaction 13(2),
73-88, 2020.
“Alexandru Ioan Cuza” University, Faculty of Computer Science
Bulevardul Carol I, Nr.11, 700506, Iaşi, România
Abstract. Mastering a musical instrument is a task that requires very much time, patience, and motivation. Many people discover their passion for music late. Even though they begin to follow a path to become great musicians or they just search for the perfect hobby, it is hard to maintain the desire to learn this skill due to the low amount of free time. Interactivity is a very important element when learning music and therefore, this paper presents a method to help new guitarists who don’t have the possibility to search for music teachers. In this paper, we will see details about Ear Trainer for Guitar Chords an Android application created to help train the ear of those who learn to play the guitar. The experiments we did showed that the application was appreciated by beginner, intermediate or advanced musicians.
Keywords: chord recognition, note recognition, discrete Fourier transform, frequency, hamming window
Cite this paper as:
Cioată, M.-A., Iftene, A. Ear Trainer for Guitar Chords – An Android Application for Recognition of Guitar Chords.
International Journal of User-System Interaction 13(2),
89-109, 2020.
1 University Politehnica of Bucharest
313 Splaiul Independentei, Bucharest, Romania
2 Institutul de Cercetări în Inteligenţa Artificială
Calea 13 Septembrie nr. 13, Bucureşti
3 Academy of Romanian Scientists
Splaiul Indpendentei 54, Bucharest, Romania
Abstract. This paper presents an approach to automatically colorize black and white images using artificial intelligence. Because different colors can be used to paint the same object, a machine learning algorithm can learn these colors during the training phase and generate realistic images. The solution is based on generative adversarial networks, a machine learning framework that uses two neural networks (the generator and the discriminator) to generate new data. The two models are trained simultaneously using the GPU support offered by Google Colab, the generator trying to deceive the discriminator with different colorizations methods and the discriminator classifying the images received from the generator as synthetic images. After the training phase, the resulting generator model can produce colorful versions of grayscale images from different datasets used as input. For a better comparison of the results, both qualitative and quantitative methods are used for the evaluation of the trained models, and also a Turing test, the algorithm obtaining a 23% chance of misleading the participants into choosing the generated image from pairs of real and synthetic images.
Keywords: image generation, artificial intelligence, machine learning, convolutional neural networks, generative adversarial network
Cite this paper as:
Petre, G.-C., Trăuşan-Matu, S. Automatic black and white image colorization using generative adversarial networks.
International Journal of User-System Interaction 13(2),
110-120, 2020.