Maider Zamalloa

Feature Selection Based on Genetic Algorithms for Speaker Recognition

The Mel-frequency cepstral coefficients (MFCC) and their derivatives are commonly used as acoustic features for speaker recognition. The issue arises of whether some of those features are redundant or dependent on other features. Probably, not all of them are equally relevant for speaker recognition. Reduced feature sets allow more robust estimates of the model parameters. Also, less computational resources are required, which is crucial for real-time speaker recognition applications using low-resource devices. In this paper, we use feature weighting as an intermediate step towards feature selection. Genetic algorithms are used to find the optimal set of weights for a 38-dimensional feature set, consisting of 12 MFCC, their first and second derivatives, energy and its first derivative. To evaluate each set of weights, speaker recognition errors are counted over a validation dataset. Speaker models are based on empirical distributions of acoustic labels, obtained through vector quantization. On average, weighting acoustic features yields between 15% and 25% error reduction in speaker recognition tests. Finally, features are sorted according to their weights, and the K features with greatest average ranks are retained and evaluated. We conclude that combining feature weighting and feature selection allows to reduce costs without degrading performance

Middleware for distributed services and mobile applications

In this paper, we describe the middleware that has evolved from our attempt to develop location-aware and user-aware services in ubiquitous environments. User mobility can cause numerous problems and limitations: user-preferred applications are not installed in the new location or device, user-specific configurations are needed, there are different OS and platforms, limited resources in the new node and so on. These restrictions sometimes prevent the user from performing what is required, anywhere and anytime. We propose a system in which, by enabling code installation and execution on remote nodes, these limitations are controlled. A prototype implementation of this approach has been constructed and we present the rationale, design, implementation and applications of the prototype system.

Low-latency online speaker tracking on the AMI Corpus of meeting conversations

Ambient Inteligence aims to create smart spaces providing services in a transparent and non-intrusive fashion, so context awareness and user adaptation are key issues. Speech can be exploited for user adaptation in such scenarios by continuously tracking speaker identity. However, most speaker tracking approaches require processing the full audio recording before determining speaker turns, which makes them unsuitable for online processing and low-latency decision-making. In this work a low-latency speaker tracking system is presented, which deals with continuous audio streams and outputs decisions at one-second intervals, by scoring fixed-length audio segments with a set of target speaker models. A smoothing technique is explored, based on the scores of past segments, which increases the robustness of tracking decisions to local variability. Experimental results are reported on the AMI Corpus of meeting conversations, revealing the effectiveness of the proposed approach when compared to an offline speaker tracking approach developed for reference.