This paper explores the application of the Short-Time Fourier Transform (STFT) for feature extraction in audio recognition, specifically using the Free Spoken Digit Dataset. The STFT is employed to analyze the time-varying spectral content of the spoken digit samples, aiming to extract discriminative features that aid in accurate digit classification. The dataset consists of recordings of spoken digits from multiple speakers, with variations in pitch, accent, and speaking style. The STFT is computed on short overlapping segments of the audio signals, providing a time-frequency representation of the data. From the STFT, magnitude spectra are obtained, capturing the energy distribution across different frequency bands over time. Several spectral features are derived from the magnitude spectra, including spectral centroid, spectral flux, and spectral roll-off. These features serve as inputs to machine learning algorithms, enabling the classification of spoken digits. Experimental results demonstrate the effectiveness of the STFT-based feature extraction approach for audio recognition. The combination of spectral features derived from the STFT achieves high classification accuracy on the Free Spoken Digit Dataset. The analysis reveals that the STFT captures important spectral patterns, facilitating discrimination between different spoken digits and mitigating the effects of variations in speakers and speaking styles.