Hi,
I need to find on certain factors that may affecting the fusion of feature vector on speech signals? Like the usage of MFCC and LPCC or other algorithm such as MFCC and MLP as bottleneck feature.
I have find out that the usage of MFCC and LPCC as input for feature vector and gives very good accuracy but when it fused together the results slightly downgraded although have tried with matric changes. Please advise.
Thank you
The number of each element in each feature vector can be a factor. Moreover, applying some feature selection and reduction techniques such as PCA, ICA, Meta heuristics (BAT, Ant, Bees, Swarm and etc) can be a good analysis over the feature vectors fusion. Furthermore, if you can remove less important features and redundant features, it can improve the performance. Good Luck
There are various optimization based feature selection techniques ,which is used to remove the redundant features which improves the accuracy.In general it is best to go for prediction of feature before feature selection.QPSO,cuckoo search are some of the optimization techniques used for the feature selection
Hi
I would suggest you to use Weka-An open source Java based software which has various feature selection and classification algorithms such as ANN, Baysian networks, SVM and so on. It is also user-friendly software tool supporting lot of features intended for machine learning. However you have got to make the dataset in CSV format in order for the weka to read and do further processing. Kindly write to me if you have any specific clarifications in using the weka tool for solving your issues.
Hope it helps!
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing