Using CRF to classify an entire sequence, not single elements

Hi all,

for my thesis I have to classify sequences of vectors (representing linguistic and sound features of spoken words) in a few number of classes (representing speech emotions), starting from some datasets of classified sequences. Now, CRF models seem to work better than HMM for labeling tasks, so it would be interesting to use them to classify the entire sequence, not single tokens.

The problem is: how can I train and use a CRF model in order to classify the entire sequence of elements? I understand that I could use a HMM for this, by creating different models for different emotions and calculating the likelihood of a sequence for each of them, but how can I use a CRF to do so? Which library do you suggest (I use Java and Python, but other languages are welcome, too) ? EDIT: punctuation