Free Spoken Digit Dataset (FSDD)

Estimated reading: 2 minutes 1648 views

Visualization of the Spoken MNIST Dataset in the Deep Lake UI

Free Spoken Digit Dataset dataset

What is Free Spoken Digit Dataset?

The Free Spoken Digit Dataset, or the Spoken MNIST (Modified National Institute of Standards and Technology database) dataset contains recordings of spoken digits in wave files at 8kHz. The dataset recordings have been trimmed at the beginning and end so that they have near-minimal silence. This dataset contains 50 recordings for every 10 digits per every 6 speakers. This dataset in total contains 3000 recordings in English pronunciations.

Download Spoken MNIST Dataset in Python

Instead of downloading the Spoken MNIST dataset in Python, you can effortlessly load it in Python via our Deep Lake open-source with just one line of code.

Load Spoken MNIST Dataset in Python

				
					import deeplake
ds = deeplake.load("hub://activeloop/spoken_mnist")

Spoken MNIST Dataset Structure

Spoken MNIST Data Fields

audio: tensor containing audio recordings.
labels: tensor containing labels for their respective audio.
spectrograms: tensor containing spectrogram of their recording.
speakers: tensor containing speaker id for audio.

How to use Spoken MNIST Dataset with PyTorch and TensorFlow in Python

Train a model on the Spoken MNIST dataset with PyTorch in Python

Let’s use Deep Lake built-in PyTorch one-line dataloader to connect the data to the compute:

				
					dataloader = ds.pytorch(num_workers=0, batch_size=4, shuffle=False)

Train a model on the Spoken MNIST dataset with TensorFlow in Python

				
					dataloader = ds.tensorflow()

Additional Information about Spoken MNIST Dataset

Spoken MNIST Dataset Description

Homepage: https://zenodo.org/record/1342401.
Repository: https://github.com/Jakobovski/free-spoken-digit-dataset.
Paper: Jackson, Z., Souza, C., Flaks, J., Pan, Y., Nicolas, H., & Thite, A. (2018). Jakobovski/free-spoken-digit-dataset: v1. 0.8.

Spoken MNIST Dataset Curators

Jackson Z, Souza C, Flaks J, Pan Y, Nicolas H, Thite A

Spoken MNIST Dataset Licensing Information

Creative Commons Attribution-ShareAlike 4.0 International

Spoken MNIST Dataset Citation Information

				
					@misc{jackson2018jakobovski,
  title={Jakobovski/free-spoken-digit-dataset: v1. 0.8},
  author={Jackson, Zohar and Souza, C{\'e}sar and Flaks, Jason and Pan, Yuxin and Nicolas, Hereman and Thite, Adhish},
  year={2018},
  publisher={Zenodo}
}

Free Spoken Digit Dataset FAQs

What are the main differences between MNIST and Free Spoken Digit dataset?

Free Spoken Digit Dataset is meant as an audio substitute for the MNIST dataset.
Both Free Spoken Digit Dataset and MNIST datasets have 10 classes: the ten digits (0 to 9) for MNIST, whilst Free Spoken Digit Dataset has the ten digits (50 of each digit per speaker, with a total of 6 speakers)