In recent years, emotion-aware systems have gained popularity in improving personalized user experiences, particularly in music therapy. Emotions and music are closely related, music has therapeutic advantages. Current algorithms for recommending music mostly depend on user behavior or textual preferences, frequently ignoring user’s emotional state, particularly as it is inferred from speech. In terms of context-aware personalization, this results in a substantial gap. We propose SERMuse: Speech Emotion Recognition (SER) based Music Recommendation. After extracting speech features using Mel-Frequency Cepstral Coefficients (MFCC), SERMuse uses a Convolutional Neural Network (CNN) to classify speech into one of the eight emotions. Next, the emotions are mapped to relevant music and content based recommender algorithm is employed. The experiments on RAVDESS dataset achieved an overall F1 score of 0.95. Using the detected emotions, emotions are mapped to data moods dataset and music recommendation system achieved an average precision of 0.60 and average NDCG of 0.72 for the top 5 recommended tracks. By incorporating emotional awareness into the music recommendation process, SERMuse helps achieve Sustainable Development Goal 3, that is, Good Health and Well-Being. It provides a non-invasive and easily available emotional support, which is especially helpful for visually impaired people going through emotional discomfort.