Machine learning has emerged as a powerful tool for audio analysis, enabling applications such as pitch detection, speech understanding, musical instrument recognition, and music generation. By transforming sound waves into spectrograms, which represent audio frequencies visually, machine learning techniques can identify intricate patterns within audio data. This research explores audio classification, specifically the differentiation between speech and music, employing Librosa, an open-source Python library for audio and music analysis. Librosa facilitates preprocessing by resampling audio to a default rate of 22,050 Hz and converting signals to mono, ensuring uniformity in data representation. Additional tools such as Scipy for feature extraction, Numpy for array processing, and Matplotlib for visualization complement this process. The audio classification task utilizes a neural network model developed with Keras and employs Sklearn for splitting data into training and testing sets. This study emphasizes the integration of these libraries and frameworks, illustrating the systematic approach required to construct robust machine learning models for audio classification.