Pasi Fränti

Real-time speaker identification and verification

In speaker identification, most of the computation originates from the distance or likelihood computations between the feature vectors of the unknown speaker and the models in the database. The identification time depends on the number of feature vectors, their dimensionality, the complexity of the speaker models and the number of speakers. In this paper, we concentrate on optimizing vector quantization (VQ) based speaker identification. We reduce the number of test vectors by pre-quantizing the test sequence prior to matching, and the number of speakers by pruning out unlikely speakers during the identification process. The best variants are then generalized to Gaussian mixture model (GMM) based modeling. We apply the algorithms also to efficient cohort set search for score normalization in speaker verification. We obtain a speed-up factor of 16:1 in the case of VQ-based modeling with minor degradation in the identification accuracy, and 34:1 in the case of GMM-based modeling. An equal error rate of 7% can be reached in 0.84 s on average when the length of test utterance is 30.4 s.

Iterative shrinking method for clustering problems

Agglomerative clustering generates the partition hierarchically by a sequence of merge operations. We propose an alternative to the merge-based approach by removing the clusters iteratively one by one until the desired number of clusters is reached. We apply local optimization strategy by always removing the cluster that increases the distortion the least. Data structures and their update strategies are considered. The proposed algorithm is applied as a crossover method in a genetic algorithm, and compared against the best existing clustering algorithms. The proposed method provides best performance in terms of minimizing intra-cluster variance.

Fast Agglomerative Clustering Using a k-Nearest Neighbor Graph

We propose a fast agglomerative clustering method using an approximate nearest neighbor graph for reducing the number of distance calculations. The time complexity of the algorithm is improved from O(tauN2) to O(tauN log N) at the cost of a slight increase in distortion; here, tau denotes the lumber of nearest neighbor updates required at each iteration. According to the experiments, a relatively small neighborhood size is sufficient to maintain the quality close to that of the full search

Outlier detection using k-nearest neighbour graph

We present an outlier detection using indegree number (ODIN) algorithm that utilizes k-nearest neighbour graph. Improvements to existing kNN distance-based method are also proposed. We compare the methods with real and synthetic datasets. The results show that the proposed method achieves reasonable results with synthetic data and outperforms compared methods with real data sets with small number of observations.

Compression of Digital Images by Block Truncation Coding: A Survey

Block truncation coding (BTC) is a lossy moment preserving quantization method for compressing digital gray-level images. Its advantages are simplicity, fault tolerance, the relatively high compression efficiency and good image quality of the decoded image. Several improvements of the basic method have been recently proposed in the literature. In this survey we will study the basic algorithm and its improvements by dividing it into three separate tasks; performing quantization, coding the quantization data and cling the bit plane. Each phase of the algorithm will be analyzed separately. On the basis of the analysis, a combined BTC algorithm will be proposed and the comparisons to the standard JPEG algoritbm will be made

Genetic Algorithms for Large-Scale Clustering Problems

We consider the clustering problem in the case where the distances between elements are metric and both the number of attributes and the number of clusters are large. In this environment the genetic algorithm approach gives high quality clusterings, but at the expense of long running time. Three new and efficient crossover techniques are introduced here. The hybridization of the genetic algorithm and k-means algorithm is discussed.

Reduced-search dynamic programming for approximation of polygonal curves

Approximation of polygonal curves with minimum error (min-e problem) can be solved by dynamic programming, or by graph-theoretical approach. These methods provide optimal solution but they are slow for a large number of vertices. Faster methods exist but they lack the optimality. We try to bridge the gap between the slow but optimal, and the fast but sub-optimal algorithms by giving a new near-optimal approximation algorithm based on reduced-search dynamic programming. The algorithm can be iterated as many times as further improvement is achieved in the optimization. It is simple, fast, and it has a low space complexity.

Randomised Local Search Algorithm for the Clustering Problem

Abstract:We consider clustering as a combinatorial optimisation problem. Local search provides a simple and effective approach to many other combinatorial optimisation problems. It is therefore surprising how seldom it has been applied to the clustering problem. Instead, the best clustering results have been obtained by more complex techniques such as tabu search and genetic algorithms at the cost of high run time. We introduce a new randomised local search algorithm for the clustering problem. The algorithm is easy to implement, sufficiently fast, and competitive with the best clustering methods. The ease of implementation makes it possible to tailor the algorithm for various clustering applications with different distance metrics and evaluation criteria.

Genetic algorithm with deterministic crossover for vector quantization

Genetic algorithm (GA) provides high quality codebooks for vector quantization (VQ) at the cost of high running time. The crossover method is the most important choice of the algorithm. We introduce a new deterministic crossover method based on the pairwise nearest neighbor method. We show that high quality codebooks can be obtained within a few minutes instead of several hours as required by the previous GA-based methods. The method outperforms all comparative codebook generation methods in quality for the tested training sets.

A fast exact GLA based on code vector activity detection

This paper introduces a new method for reducing the number of distance calculations in the generalized Lloyd algorithm (GLA), which is a widely used method to construct a codebook in vector quantization. Reduced comparison search detects the activity of the code vectors and utilizes it on the classification of the training vectors. For training vectors whose current code vector has not been modified, we calculate distances only to the active code vectors. A large proportion of the distance calculations can be omitted without sacrificing the optimality of the partition. The new method is included in several fast GLA variants reducing their running times over 50% on average.