Speech emotion recognition is a paramount topic due to its applications in affective computing, marketing, and mental health. Although extensive research has been conducted on SER for languages such as English and Chinese, studies in Spanish remain scarce despite Spanish being one of the most spoken languages in the world. In particular, there is a lack of research on deep learning techniques for SER in Spanish. This paper proposes a method based on the logarithmic mel spectrogram along with a deep neural network for emotion recognition from Spanish speech. The selection of the neural network is based on evaluating the performance and computational cost of models commonly used in emotion recognition, including ResNet-50, MobileNet-V2, EfficientNet-B0, VGG-16, Inception-V3, and Long Short-Term Memory networks (LSTMs). The experiments were conducted using the EmoMatchSpanishDB dataset, which consists of 2,005 Spanish speech signals classified into seven emotion categories. The results indicate that the proposed method performs best with EfficientNet-B0 and outperforms other methods tested on the EmoMatchSpanishDB dataset.