Speech Commands Dataset Enhanced for Direction-of-Arrival Estimation

This dataset can serve as a training and evaluation corpus for the task of training keyword detection with speaker direction estimation (keyword direction of arrival - KWDOA). It was created by processing the existing Speech Commands dataset [1] with the PyroomAcoustics library so that the resulting speech recordings simulate the usage of a circular microphone array with 4 microphones having a distance of 57 mm between adjacent microphones. Such design of a simulated microphone array was chosen in order to match the existing physical microphone array from the Seeeduino series. [1] Warden, Pete. “Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition.” ArXiv.org, 2018, arxiv.org/abs/1804.03209