RuntimeError: logadd: minusdif
Created by: Artimi
Hi,
I'm using bob 1.2.2 in xubuntu VM. I try to train GMM with ML on MFCCs as it is described in documentation. First use kmeans.means, then create GMMMachine, create ML_GMMTrainer and call train on all of that. But while it is training I get Runtime Error:
Traceback (most recent call last):
File "./speech.py", line 248, in <module>
trainer.train(args.train_machine)
File "./speech.py", line 190, in train
gmm = self.train_machine(class_number)
File "./speech.py", line 169, in train_machine
trainer.train(gmm, data)
RuntimeError: logadd: minusdif (-nan) log_b (-nan) or log_a (-22.929673) is nan
Basically I have same problem as was discussed here https://groups.google.com/forum/#!topic/bob-devel/VOi8k0Ts1gw, but it did not help when I set kmeansTrainer.initialization_method to KMEANS_PLUS_PLUS or RANDOM_NO_DUPLICATE.
You can see current code here https://github.com/Artimi/stws2014/blob/master/code/speech.py