The recognition of emotions in speech is one of the most challenging topics in data science. In this work, we define a pipeline for the study of multimodal speech recognition, using a wide set of features from audio samples and text transcripts. This work aims to study the interaction and contribution of multimodal features and for this purpose, three types of features have been selected. We extract a set of handcrafted features related to speech prosody, along with classical mel spectrogram acoustic features and TF-IDF for text. Combining these three types of data we evaluate the contribution that they represent to each other. This Thesis also provides a comparative study between the classical machine learning models performance over neural architectures in terms of performance and learning potential from speech. Finally, it presents an application that provides emotion classification and feedback retrieval for misclassified samples.