I'm making a program emotion-recognition in speech.
Using a mfcc + K-mean + GMM, I got a likelihood like -15012, -43400, -8000
It was so bigger value than I though. Is it right value?
Wave file's length usually 2~3 and I init using below option.
MFCC option
SampleRate = 16000
WindowSize = 256
NumberCoefficients = 13
UseFirstCoefficient = true
MinFreq = 20.0
MaxFreq = 16000.0
NumberFilters = 25
K-Means Clustering default option
NumberCluster = 2
The actual values of the likelihoods depend on many parameters used during feature extraction. For example, filterbank might be created in a different ways. They are different across different engines. So just looking on the values you can not say if they are correct or not.
Since you didn't mention the engine it's not possible to answer this question.