@inproceedings {pub3081,
	title = {Speaker Adaptation for Word Prominence Detection with Support Vector Machines},
	author = {Andrea Schnall AND Martin Ernst Heckmann},
	year = {2016},
	month = {May},
	abstract = {In this paper we propose a new adaptation method to improve the detection of prominent from non-prominent words. Prosodic cues are difficult to extract, due to the different features different speakers are using to express for example prominence in speech. To overcome the problem of variations from the pool of speakers used during training time and those encountered during deployment, 
in speech recognition speaker-adaptation techniques like  feature-space Maximum Likelihood Linear Regression (fMLLR) turned out to be very useful. In the case of prominence detection, our former results showed, that a discriminative classifier like the support vector machines (SVM) works better than GMM-HMM based classifiers. Since existing adaptation methods like fMLLR are developed for GMM-HMM based classifiers and the assumption that the data has a Gaussian distribution does not hold for our data, using the fMLLR with the SVM leads not to an improvement for our problem case. 
Therefore we propose a new adaptation method, which combines the fMLLR with an adaptation to the radial basis function kernel of the SVM. We investigate how this method can be used to adapt a new speaker to a speaker independently trained model for word prominence detection. We show that the performance improves from the speaker adaptation from 16.4\% error rate to 14.4\%. },
	publisher = { Speech Prosody 2016},
	booktitle = { Speech Prosody 2016}
}