Google Speech Commands Dataset

The Google Speech Commands Dataset was created by the TensorFlow and AIY teams to showcase the speech recognition example using the TensorFlow API. The dataset has 65,000 clips of one-second-long duration. Each clip contains one of the 30 different words spoken by thousands of different subjects. 

Note

The Google Speech Commands Dataset is available from the following link: http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz.

The clips were recorded in realistic environments with phones and laptops. The 35 words contained noise words and the ten command words most useful in a robotics environment, and are listed as follows:

• Yes
• No
• Up
• Down
• Left
• Right
• On
• Off
• Stop
• Go

More details on how the speech dataset is prepared can be found in the following links:

• https://arxiv.org/pdf/1804.03209.pdf
• https://ai.googleblog.com/2017/08/launching-speech-commands-dataset.html

With this dataset, thus the problem that shown in the example in this chapter is known as Keyword Spotting...