Speech emotion recognition algorithm targets to detect the speaker’s emotional states from speech. In speech emotion recognition task, Support Vector Machine, Gaussian Mixture Model, Hidden Markov Model and Neural Networks are the most popular classification algorithms. Analysis and comparison influence of different types of spectral and prosodic features for emotional speech classification based on Gaussian Mixture Models (GMM). Influence of initial setting of parameters (number of mixture components and used number of iterations) for GMM training process was also analyzed. Subsequently, analysis was performed to find how correctness of emotion classification depends on the number and the order of the parameters in the input feature vector and on the computation complexity . The databases available are in German, English, Swedish, Mandarin and Indian languages such as Hindi and Tamil. This speech addresses an important aspect regarding speech emotion recognition of normal and autistic children. In addition, it will also discuss to develop a procedure to help the psychologist to identify the intensity of autism in spoken utterances of children. The process consists of three parts. The first part investigates the best feature to do the analysis. The features are extracted then threshold is applied for every feature to develop an algorithm. The second part comprises of implementing an algorithm which serves as a classifier for normal and autistic speech. The third part comprises of comparing the developed algorithm with traditional machine learning algorithms to analyze the efficiency of the developed algorithm. The research statements that were carried out previously put a great focus on recognition of speech emotions in different areas and the effect of different speech features used in emotion recognition. The comparative analysis of speech emotion for normal and Autistic children in Urdu language using learning classifiers has not been extensively investigated. This speech focuses on the implementation of the algorithm that serves as the classifier for the classification of the emotions of normal and Autistic children in Urdu Language. Detecting the emotions of the autistic children is not that easy as it seems to be, that is why learning techniques are used in this study to overcome this research gap. The methodology of this study comprises on 1) extraction of speech features both prosodic and spectral, 2)implementation and testing of the algorithm that classify the speech emotions of normal and autistic children based on classification results of Berlin Emotional Database (EMO DB) in comparison with results of proposed classifier SERNAC (Speech Emotion Recognition of Norma and Autistic Children) in Urdu Language, and 3) improvement of the classification accuracy by analyzing the extracted reduced feature sets on the proposed algorithm.