Abstract:In this paper we present a method for algorithmic melody generation using a generative adversarial network without recurrent components. Music generation has been successfully done using recurrent neural networks, where the model learns sequence information that can help create authentic sounding melodies. Here, we use DC-GAN architecture with dilated convolutions and towers to capture sequential information as spatial image information, and learn long-range dependencies in fixed-length melody forms such as Irish traditional reel.
Abstract:This paper describes the Hangulphabet, a new writing system that should prove useful in a number of contexts. Using the Hangulphabet, a user can instantly see voicing, manner and place of articulation of any phoneme found in human language. The Hangulphabet places consonant graphemes on a grid with the x-axis representing the place of articulation and the y-axis representing manner of articulation. Each individual grapheme contains radicals from both axes where the points intersect. The top radical represents manner of articulation where the bottom represents place of articulation. A horizontal line running through the middle of the bottom radical represents voicing. For vowels, place of articulation is located on a grid that represents the position of the tongue in the mouth. This grid is similar to that of the IPA vowel chart (International Phonetic Association, 1999). The difference with the Hangulphabet being the trapezoid representing the vocal apparatus is on a slight tilt. Place of articulation for a vowel is represented by a breakout figure from the grid. This system can be used as an alternative to the International Phonetic Alphabet (IPA) or as a complement to it. Beginning students of linguistics may find it particularly useful. A Hangulphabet font has been created to facilitate switching between the Hangulphabet and the IPA.