數位音頻分析系統研究-以應用於KTV評分

Abstract

本研究旨在以類神經網路訓練技術模擬人工歌唱評審程序作為卡拉OK系統評分技術，發展目前流行之卡拉OK設備評分系統之不足。 卡拉OK評分系統中，最直接的就是對評分方式的取決，本文將傳統評分方式由音量與音軌是否符合原唱者的差異性來評分，改由類神經訓練與模擬人工評審評分程序，增加歌唱者原始之聲音歌唱特性，使得原有之評分系統更具客觀與公正性。 實驗之環境完全比照目前卡拉OK歌唱之環境，類神經訓練之樣本完全由此實務環境擷取出來，故實驗之結果更具有參考價值

The purpose of this thesis is to develop a better score system for the Karaoke singers from the technology of ‘Neural Network Training’ and the simulation of the judging process of human juries. Several ways of scoring are developed under different Karaoke score system. The traditional way of score system is based on comparing the volumes and the tracks of the original singers with those of the Karaoke singers. The problems of lacking objectivity and fair have been identified. In order to rectify these problems, this thesis adopts the alternative approach. This newer and better way of scoring is based on the technology of neural network training and the simulation of the judging process from human juries and is able to consider the singing and vocal characteristics of the original singers in a better way. The samples of the neutral network training are all gotten under the practical environment of Karaoke singing. Thus, these experiments are considered as the valuable contributions of this thesis.