音樂情緒分析系統於 Android 系統之實作研究

Abstract

多媒體裝置的普及讓音樂檔案易於紀錄儲存，然而資料量龐大的音樂資料庫，讓使用者管理與檢索困難，如何讓電腦有效率並且讓使用者能夠人性化管理音樂資料庫，是現今音樂資料檢索(Music Information Retrieval)的研究方向。以往對於音樂資料庫是由曲目、作者、專輯、音樂類型…等各種文字資訊作為檢索指標，近期有越來越多相關研究使用以不同資訊作為檢索方法，又以基於內容的音樂檢索(Content-based Music Information Retrieval)為研究大宗，而基於內容的音樂檢索是利用音樂本身特徵(features)方法來作為音樂檢索的依據。本論文研究目標藉由分析五種音樂的特徵方法，音樂事件密集度 (Onset Density) 、音量 (Volume) 、音色 (Timbre) 、調性 (Mode) 、不和諧度(Dissonance)，計算其特徵值與聆聽者情緒相關之壓力-能量數值，並經由 200 首樂曲之分類訓練，完成音樂情緒分析，並將此系統系統實作於 ASUS Transformer Pad Infinity (TF700T)平板電腦上，藉由其強大的運算能力實現於 Android 可攜帶式智慧型行動裝置上。本系統可於行動裝置上顯示其音樂資料庫中樂曲之情緒成分，藉此展現音樂資料庫之情緒分布。

The popularity of electronic multimedia devices has made it easy to record and store the music contents. However, the huge amount of music information database increases the difficulties of users to manage and retrieve the music contents. The current research in Music Information Retrieval aims to make computer works efficiently and also allows a more user-friendly music library for managing music contents. The music library in the past are mainly using the song titles, artists, albums, music genres and other text information as the search index. Nowadays, there are a growing number of recent studies which attempt to use different kinds of information as the retrieval method. These studies are based on the Content-Based Music Information Retrieval which utilizes the music features as the basis for music retrieval. The objective of this thesis is to implement the musical mood analysis system on the Android tablet computer: ASUS Transformer Pad Infinity (TF700T) by analyzing the five music features: Onset Density, Volume, Timbre, Mode and Dissonance with its powerful computing capabilities and this has enabled the musical mood analysis system to run on the Android mobile devices.