Characteristic analysis of auditory perception and aesthetics in sound composition optimization using revised interactive differential evolution
| Autor: | Hayato Shindo, Yan Pei |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Auditory perception
education.field_of_study Computer science Speech recognition Population Crossover k-means clustering Initialization Interactive evolutionary computation 0102 computer and information sciences 02 engineering and technology 01 natural sciences 010201 computation theory & mathematics Differential evolution 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing education Timbre |
| Zdroj: | GECCO Companion |
| Popis: | In this study, we analyse auditory perception and sound aesthetic characteristic using a sound composition optimizations system, which is designed and implemented using a revised interactive differential evolution algorithm. We revise the population initialization and crossover operation of a conventional interactive differential evolution algorithm. The initialization population of the sound composition system uses a set of well-designed sounds as individuals rather than these randomly generated in a search space. The crossover operation of a conventional interactive differential evolution algorithm is revised in order to maintain the search space within a feasible range. The differential mechanism of interactive differential evolution provides a more feasible combination of these well-designed sounds to create a new type of sounds in the timbre domain. The subjective evaluation of revised interactive differential evolution algorithm leads to the sound composition to satisfy the preference of a subject as much as possible. We invite five subjects to conduct the experimental evaluation to create sounds with their personal preferences. We analyse the sound aesthetic characteristic of these five subjects using principal component analysis and k-means clustering methods with the sound data selected in the optimization process. The characteristics of selected sound help to explain optimization process and convergence of the algorithm, and assist to better understand the aesthetic judgement of each subject. We found that the proposed system and method can generate or create a sound satisfying a subject's preference. The proposal is an efficient method to analyse the characteristics of auditory perception and sound aesthetic of the subjects as well. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|