This study presents a multi-view learning framework for speech emotion recognition (SER) based affective profiling in Indian speech data by introducing analytic signal matched filter bank (ASMFB) features. We explore categorical emotion classification and dimensional emotion representation using the valence–arousal–dominance (VAD) model. A lightweight convolutional neural network (CNN) is trained on features extracted using ASMFB, which captures time–frequency information relevant for emotional expression in speech. The model predicts one of five emotion classes–happy, sad, fear, surprise, and neutral, and these are then mapped to fixed VAD values derived from standardized psychological resources. To ensure cultural and acoustic relevance, we use the Indian Emotional Speech Corpus, a dataset where speakers express different emotions using the same utterances. The framework supports both discrete and continuous affect modeling and includes visualizations such as 2D and 3D VAD scatter plots and emotional trajectory curves over time. By integrating emotion classification and VAD mapping in a single pipeline, this work offers a practical and interpretable approach to SER for Indian speech, with potential applications in affective computing, emotion-aware systems, and real-world audio analytics. Our model achieved a classification accuracy of 89.87% and an average VAD distance (error) of 0.0308, showing a strong ability to accurately recognize the emotional tone of speech based on arousal, valence and dominance. Compared to previous studies, this result marks a clear improvement over earlier studies. These findings suggest that our approach is not only reliable but also more precise in modeling the subtle emotional dimensions present in human speech.