Article details

Research area
Speech recognition

Interspeech 2012, At Portland, OR, USA


Raymond Brueckner, Björn Schuller

Likability Classification – A Not So Deep Neural Network Approach


This papers presents results on the application of restricted Boltzmann machines (RBM) and deep belief networks(DBN) on the Likability Sub-Challenge of the Interspeech 2012 Speaker Trait Challenge [1]. RBMs are a particular form oflog-linear Markov Random Fields and generative models whichtry to model the probability distribution of the underlying inputdata which can be trained in an unsupervised fashion. DBNscan be constructed by stacking RBMs and are known to yieldan increasingly complex representation of the input data as thenumber of layers increases. Our results show that the Likability Sub-Challenge classification task does not benefit from the modeling power of DBN, but that the use of an RBM as the first stage of a two-layer neural network with subsequent fine-tuning improves the baseline result of 59.0 % to 64.0 %, i.e., a relative 8.5 % improvement of the unweighted average evaluation measure.

Read/download now