Fact · Speech recognition
2048units per hidden layer
In the 2012 Switchboard experiment, a seven-hidden-layer deep network with 2,048 units in each layer replaced the Gaussian mixture model in the acoustic model.
Checked against its source on October 17, 2026
In the source's own words
“A seven-hidden-layer DBN-DNN with 2,048 units in each layer and full connectivity between adjacent layers replaced the GMM in the acoustic model.”
Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups, Geoffrey Hinton et al., IEEE Signal Processing Magazine, 2012, Switchboard speech recognition task.
Open the original (DOI 10.1109/MSP.2012.2205597)- What it measures
- The network was trained on the 309 hours of the standard Switchboard-I training set, as described in the same section.
- What it does not show
- One system in one experiment, described by its authors. Later networks differ in size and design.