We current Spatial LibriSpeech, a spatial audio dataset with over 570 hours of 19-channel audio, first-order ambisonics, and optionally available distractor noise. Spatial LibriSpeech is designed for machine studying mannequin coaching, and it contains labels for supply place, talking route, room acoustics and geometry. Spatial LibriSpeech is generated by augmenting LibriSpeech samples with >220k simulated acoustic circumstances throughout >8k artificial rooms. To reveal the utility of our dataset, we prepare fashions on 4 elementary spatial audio duties, leading to a median absolute error of 6.60° on 3D supply localization, 0.43m on distance, 90.66ms on T30, and a couple of.74dB on direct-to-reverberant ratio estimation. We present that the identical fashions switch to widely-used analysis datasets, acquiring, as an example, a median absolute error of 12.43° on 3D supply localization on TUT Sound Occasions 2018, and 157.32ms on T30 estimation on ACE Problem.