Feature Selection Based on Genetic Algorithms for Speaker Recognition

Feature selection algorithms for automatic speech recognition

“…Candidate solutions are represented by individuals (or chromosomes) in a large population. Initial solutions may be randomly generated or obtained by other means [1]. Then GAs iteratively drive the population to an optimal point according to a complex metric (called fitness or evaluation function) that measures the performance of the individuals in a target task.…”

Detailed analysis of Speaker Recognition System and use of MFCCs for recognition

“…[2] For feature set reduction many methods are suggested because on account of identification of a particular speaker we get a lot of features which need to be reduced to only the features capable of representing someone PCA(Principal Component Analysis) is one such technique where the eigenvectors are calculated which are then sorted in descending order and a projection matrix is built finally known as the Karhunen-Loeve Transforn(KLT)with the largest K eigenvectors.KLT decorrelates the features and gives the smallest possible reconstruction error among all linear transforms, i.e. the smallest possible mean-square error between the data vectors in the original D-feature space andthe data vectors in the projection K-feature [1] Linear Discriminant Analysis (LDA) attempts to find the transform A that maximizes a criterion of class separability.This is done by computing the within-class and between class variance matrices, W and B, then finding the eigenvectors of W and B, sorting them according to the eigen values, in descending order, and finally building the projection matrix A with the largest K eigenvectors (which define the K most discriminative hyperplanes). LDA assumes that all classes share a common within-class covariance, anda single Gaussian distribution per class.…”

“…Therefore it is always wiser to go for low level features because not only they can be extracted easily but they do not require a lot of data. In most of the speech and speaker recognition systems MFCC are used because of the fact that these features apart from identifying the frequency distribution also tells the glottal sources and the vocal tract shape and length, which are the features specific to a speaker [1].Speaker recognition is being used a biometric in various security applications where the people are recognized on the basis of their voice.The main tedious task in the speaker recognition is the enrollment phase where the users are asked to input their voice via a microphone or any input device which is further used as a database in the recognition phase for identification of the correct speaker.There are some other techniques also which can be used apart from enrollment phase because it is not always possible to get inputs from various speakers.…”

Efficient and accurate image alignment using TSK-type neuro-fuzzy network with data-mining-based evolutionary learning algorithm

“…Because the 128-elemet feature vector is still too high to train a TNFN, there is a requirement of finding a dimensionality reduction method to lower the dimension of the feature vector. According to [21], genetic algorithm outperformed than principal component analysis and linear discriminate analysis as dealing with their speaker recognition case. Thus, in our image alignment case, we adopted genetic algorithm method described in [22] to reduce a 128-elemet into a 33-element feature vector in the experimental section.…”