本研究提出一個使用信號頻譜減法增強技術和類神經網路為架構的音樂資料庫的歌手識別分類系統，並期望相關技術可以被應用於車輛之多媒體系統上。信號頻譜減法增強法為一種信號處理技術，廣泛用於消除聲音訊號中的背景噪音。在本研究中，頻譜減法技術應用在含有音樂伴奏為背景噪聲條件下的歌手語音信號增強。在實驗中評估所提出的語音增強算法，頻譜減法、多頻帶頻譜減法和非線性頻譜減法三種不同方法的使用和比較。且於信號分類的步驟，將擷取出的特徵信號作為輸入向量，再以廣義回歸類神經網路（GRNN）進行歌手分類鑑定。實驗結果顯示，本研究所提出的方法對於歌手識別是有效的，也可作為相關後續研究的基礎。

This paper presents a study of a singer identification system using the signal spectral subtraction enhancement technique and artificial neural network. Signal spectral subtraction enhancement is a signal processing technique widely used for eliminating the background noise from sound signals. In the present study, the spectral subtraction technique is applied to the singer signal enhancement with background music. To compare the proposed speech enhancement algorithm, spectral subtraction, multi-band spectral subtraction and non-linear spectral subtraction are used and compared in the experimental investigation. In the signal classification stage, the post-processing features of signals are used as input information to generalize a regression neural network (GRNN) classifier for singer identification. The experimental result indicates the proposed singer identification is effective and has satisfactory performance.