پديد آورندگان :
نيك نژاد، علي دانشگاه آزاد تهران جنوب، تهران , غفاري، علي دانشگاه آزاد اسلامي واحد تهران جنوب - دانشكده فني و مهندسي، تهران، ايران , خداياري، عليرضا دانشگاه ازاد اسلامي واحد پرديس - گروه مهندسي مكانيك، تهران، ايران
كليدواژه :
تشخيص احساس , ضرايب كپسترال فركانسي مل , احساس گفتاري , ويژگي هاي صدا
چكيده فارسي :
آسان‌ترين ارتباط بين انسان و ماشين از طريق گفتار است و از ملزومات اين ارتباط، درك احساس انسان توسط ماشين است. در الگوريتم پيشنهادي، با هدف افزايش سرعت و دقت در تشخيص، از ويژگي‌هاي سيگنال صدا، ضرايب كپسترال فركانسي مل را استخراج كرده و ويژگي‌هايي بهينه انتخاب مي‌شوند. سپس با استفاده از تركيب طبقه‌بندهاي ماشين‌بردار پشتيبان و مدل مخلوط گاوسي، تشخيص احساس انجام مي‌شود. نتايج حاصل از پياده‌سازي اين الگوريتم براي پايگاه‌داده آلماني 89% و براي پايگاه‌داده فارسي 68% بدست‌آمده. مقايسه نتايج اين الگوريتم با الگوريتم‌هاي مشابه عملكرد مناسب الگوريتم را در طراحي سيستم‌هاي كنترل و هدايت ربات‌ها نشان مي‌دهد.
چكيده لاتين :
The easiest method of communication between humans and machines is through speech and one of the essentials aspects of this relationship, is perception of humanistic sentiments by machine. As a result, getting speech’s patterns and creating a system based on this model has been a challenge for researchers in recent years. Although the emotion shown in speech could ranges in a very divergent spectrum, because of pander to accent, culture and environment, but a fixed patterns could be found in feelings of people’s speech. In this paper, a new algorithm to detect emotion in human voice is presented. In proposed algorithm, features are extracted from the audio signal, inspired by human hearing. and then the optimal features are chosen from the extracted features with the aim of increasing the speed and accuracy of diagnosis with an intelligent method. In addition, classifying is done by combining a set classifiers subsequently, patterns of anger, joy, fear, boredom, disgust and sadness is distinguished by the designed intelligent system. Results of the simulations of the implemented algorithm is presented with two databases, Farsi and Germany and then compared with the outcomes of other algorithms with the same databases. Results indicate that the proposed algorithm could predict emotions of anger, joy, fears, boredom, disgust and sadness with good accuracy. This algorithm could be used in designing control systems and robot guidance. In addition, emotion recognition system could be utilized in psychology, medicine, and behavioral science and security applications such as polygraph.
