free-spoken-digit-dataset

Why we included this project

Free Spoken Digit Dataset is the small, well-understood entry point for audio classification that people keep coming back to. It follows the MNIST pattern: 3,000 trimmed 8kHz wav files from six speakers, with filenames that make the digit, speaker, and index obvious at a glance and metadata covering gender and accent. The repo bundles a silence trimmer, a small Python class for loading the data, and a spectrogram generator, and it documents a fixed train/test split so results stay comparable. A Zenodo DOI and an open contribution policy keep the dataset versioned and growing, which is why it shows up so often in courses and papers. It loads cleanly through the Activeloop Hub and TensorFlow Datasets, so teams prototyping a speech pipeline or teaching audio ML get a reproducible sandbox before moving to larger corpora.