Abstarct: Abstract Different emotional states such as anger, happiness, hatred, fear and sadness cover a major part of human life. In particular, the way a person speaks changes depending on his emotional state. Therefore, the human speech signal includes information related to the speaker's feeling or state, in addition to conveying the message. Emotional states and environmental conditions such as noise cause changes in speech parameters and reduce the efficiency of automatic speech processing (ASR) systems. This problem limits the use of these systems. In this regard, in this thesis, a feature extraction method baxsed on the AM-FM model is proposed to detect emotion in noisy conditions. The classifier used is nearest neighbor (KNN) with K value of 25. This proposed model has been evaluated in two different environmental conditions, including clean and noise-free conditions and noisy conditions. In these evaluations, the emotional data of the German language named EMO-DB has been used for speech data. The evaluations have shown that the accuracy of the proposed model was 56%. In comparison with the basic systems, it does not perform better on its own, and in order to reduce the error rate in emotional speech recognition, it is better to combine it with the basic feature extraction system.