陪伴機器人之基於情感辨識音樂播放器系統

Abstract

現今低資源需求與平價之嵌入式系統已普遍應用於各式產品中。透過使用者情感表達，可以增進人們與科技產品間之互動。本研究嘗試提出互動應用的人機介面設計，利用聲音訊號作為情感辨識基礎，進而創造一個可感知情緒的音樂播放器，以應用於智慧型手機與智慧機器人。由於人類具有複雜的情緒，例如快樂或高興，不能僅藉由某些特定的數據類別來加以區分，因此我們提出的系統可以將簡短的語音投射至二維Valence - Arousal情緒座標，使用者任何情感可以以相隊應之連續值表示。本論文之方法允許系統自動從歌曲資料庫中選取音樂，這些被選取歌曲也以二維Valence - Arousal數值呈現。此外，我們提出一個激勵策略，如果使用者聲音被偵測為負面情緒，即會自動選取並播放多樣化音樂類型，用於提升使用者情感進而達到更平靜及快樂的境界。此概念已被應用現於流行且平價嵌入式平台 Beagleboard XM，它提供使用者電子麥克風及觸控式螢幕作為輸入設備。本論文提出之系統可被使用於人機介面應用，舉凡陪伴式機器人、汽車音響及通訊設備。其中通訊設備，如手機，可以根據使用者情緒狀態播放合適音樂。

Currently resource constrained and inexpensive embedded systems have become virtually an ubiquitous technology. The way in that people interact with this technology can be improved by using emotional information from the user. In this work, a novel human-machine interaction system is proposed where emotional recognition from the speech signal is used to create an emotion aware music player that can be implemented on a standard smartphone-like embedded platform. Since a person’s emotion cannot always be classified in a certain number of categories such as happy or angry, the proposed system maps an inputted short speech utterance to a two dimensional emotional plane of valence and arousal, where any given emotions can have continuous values. This strategy allows the system to automatically select a piece of music from a database of songs, in which emotions are also expressed using arousal and valences values. Furthermore, a cheer-up strategy is proposed where in case the detected emotional content is detected as negative, music songs with varying emotional content are played in order to cheer-up the user to a more neutral –happy state. The system has been implemented in a popular inexpensive embedded platform Beagleboard XM which uses an electret microphone and a touch screen panel as input from the user. The proposed system can be used in human-machine interface applications like companion robots, car sound systems and communication devices like cellphones, where music can be played according to user’s emotional state.