Abstract: The primary objective of the proposed work is to recognize emotions through speech using Recurrent-Neural-Networks and suggest physical exercises accordingly for the beneficiary of people who prefer to choose a balanced life style in an efficient way. Recurrent-Neural-Network (RNN) is a classifier technique which is utilized to classify various emotions which are happiness, disgust, sadness, fear, surprise and anger. Culmination of multiple features on similar databases are compared and clearly discussed. The outcome gives good result. At the outset the experimental results would shows that combining feature will yield higher accuracy rate on emotional databases using RNN classifier. Bayes Classifier, Hidden Markov Model (HMM), Support Vector Machine (SVM), Kernel deterioration and K Nearest Neighbors approach (KNN), Gaussian Mixture Model (GMM), Naïve-Bayes-classifier(NBC) were classifiers used by many researchers for human emotion recognition and translation. The main disadvantage is that only single input and corresponding output is given and combinations of output is a tough task and the algorithm gets costly. Our proposed system is used to suggest the physical exercise from speech. Energy, format, pitch, few spectrum features Mel Frequency Ceptral Coefficients (MFCC) and spectral Modulation features are the various common features extracted and used in modern research. In this work to extract emotional features the modulation spectral features used. To analyze, recognize and classify the emotions from audio speech samples the Standard English emotional database are used. This system will be helpful to people who need guidance in recognizing their emotion and to get a better physical exercise suggestion.