Downloads the Free Spoken Digits dataset and loads its metadata as pandas dataframes. The audio samples are as .wav files.

import dataget

df =
Dataget doesn't load the audio dataset into memory, instead the df dataframe has the audio_path column which contains the relative path of each sample. You can easily load them using


Its recommended that you split train / test based on user instead of randomly to avoid testing based on similar samples found in training.


type shape
df pd.DataFrame (2_000, 4)


column type description
audio_path str Relative path of the audio file
label int64 Target label in the range [0, 9]
user str Name of the speaker
repetition int64 Repetition number for each (user, label) pair, i.e. each user repeats each digit multiple times


  • Folder name: audio_free_spoken_digit
  • Size on disk: 26MB