Enrique M. Albornoz

Spoken emotion recognition using hierarchical classifiers

The recognition of the emotional state of speakers is a multi-disciplinary research area that has received great interest over the last years. One of the most important goals is to improve the voice-based human-machine interactions. Several works on this domain use the prosodic features or the spectrum characteristics of speech signal, with neural networks, Gaussian mixtures and other standard classifiers. Usually, there is no acoustic interpretation of types of errors in the results. In this paper, the spectral characteristics of emotional signals are used in order to group emotions based on acoustic rather than psychological considerations. Standard classifiers based on Gaussian Mixture Models, Hidden Markov Models and Multilayer Perceptron are tested. These classifiers have been evaluated with different configurations and input features, in order to design a new hierarchical method for emotion classification. The proposed multiple feature hierarchical method for seven emotions, based on spectral and prosodic information, improves the performance over the standard classifiers and the fixed features.

Spoken Emotion Recognition Using Deep Learning

Spoken emotion recognition is a multidisciplinary research area that has received increasing attention over the last few years. In this paper, restricted Boltzmann machines and deep belief networks are used to classify emotions in speech. The motivation lies in the recent success reported using these alternative techniques in speech processing and speech recognition. This classifier is compared with a multilayer perceptron classifier, using spectral and prosodic characteristics. A well-known German emotional database is used in the experiments and two methodologies of cross-validation are proposed. Our experimental results show that the deep method achieves an improvement of 8.67% over the baseline in a speaker independent scheme.

Emotion Recognition in Never-Seen Languages Using a Novel Ensemble Method with Emotion Profiles

Over the last years, researchers have addressed emotional state identification because it is an important issue to achieve more natural speech interactive systems. There are several theories that explain emotional expressiveness as a result of natural evolution, as a social construction, or a combination of both. In this work, we propose a novel system to model each language independently, preserving the cultural properties. In a second stage, we use the concept of universality of emotions to map and predict emotions in never-seen languages. Features and classifiers widely tested for similar tasks were used to set the baselines. We developed a novel ensemble classifier to deal with multiple languages and tested it on never-seen languages. Furthermore, this ensemble uses the Emotion Profiles technique in order to map features from diverse languages in a more tractable space. The experiments were performed in a language-independent scheme. Results show that the proposed model improves the baseline accuracy, whereas its modular design allows the incorporation of a new language without having to train the whole system.

Automatic classification of Furnariidae species from the Paranaense Littoral region using speech-related features and machine learning

Over the last years, researchers have addressed the automatic classification of calling bird species. This is important for achieving more exhaustive environmental monitoring and for managing natural resources. Vocalisations help to identify new species, their natural history and macro-systematic relations, while computer systems allow the bird recognition process to be sped up and improved. In this study, an approach that uses state-of-the-art features designed for speech and speaker state recognition is presented. A method for voice activity detection was employed previous to feature extraction. Our analysis includes several classification techniques (multilayer perceptrons, support vector machines and random forest) and compares their performance using different configurations to define the best classification method. The experimental results were validated in a cross-validation scheme, using 25 species of the family Furnariidae that inhabit the Paranaense Littoral region of Argentina (South America). The results show that a high classification rate, close to 90%, is obtained for this family in this Furnariidae group using the proposed features and classifiers.

Feature extraction based on bio-inspired model for robust emotion recognition

Emotional state identification is an important issue to achieve more natural speech interactive systems. Ideally, these systems should also be able to work in real environments in which generally exist some kind of noise. Several bio-inspired representations have been applied to artificial systems for speech processing under noise conditions. In this work, an auditory signal representation is used to obtain a novel bio-inspired set of features for emotional speech signals. These characteristics, together with other spectral and prosodic features, are used for emotion recognition under noise conditions. Neural models were trained as classifiers and results were compared to the well-known mel-frequency cepstral coefficients. Results show that using the proposed representations, it is possible to significantly improve the robustness of an emotion recognition system. The results were also validated in a speaker-independent scheme and with two emotional speech corpora.

Multiple feature extraction and hierarchical classifiers for emotions recognition

The recognition of the emotional states of speaker is a multi-disciplinary research area that has received great interest in the last years. One of the most important goals is to improve the voiced-based human-machine interactions. Recent works on this domain use the proso-dic features and the spectrum characteristics of speech signal, with standard classifier methods. Furthermore, for traditional methods the improvement in performance has also found a limit. In this paper, the spectral characteristics of emotional signals are used in order to group emotions. Standard classifiers based on Gaussian Mixture Models, Hidden Markov Models and Multilayer Perceptron are tested. These classifiers have been evaluated in different configurations with different features, in order to design a new hierarchical method for emotions classification. The proposed multiple feature hierarchical method improves the performance in 6.35% over the standard classifiers.

Development and evaluation of an automatic software for management zone delineation

The lack of availability of user-friendly and automatic software for management zone delineation is limiting the adoption of site-specific management practices. Several procedures for management zone delineation have been proposed, but they commonly require the use of different software, or advanced GIS and statistical skills of users, which limit their adoption. This study proposes a user-friendly and automatic software that would integrate all steps in order to delineate management zones and make prescription files. The software includes importation of different input data layers, re-projection and resizing data in a common grid size. An integrative index was proposed for the selection of the optimal number of zones after clustering analysis. Users are guided by graphical windows showing intermediate results. Also, additional automatic post-processing techniques to improve size, shape and fragmentation of delineated zones are available. The final step allows generation of the ESRI Shapefile required to make variable rate prescriptions by zone with minimal user intervention. The performance of the approach was evaluated for management zone delineation using single and multiple layers of data by comparing with Management Zone Analyst software, and the improvement of the approach in the selection of the optimal number of zones and reducing zone-fragmentation was shown. The software design includes a simple graphical user interface and requires minimal user intervention in order to assist the end-user. The main contribution of this work was the successful development of this automatic user-friendly solution that includes all the necessary steps for management zone delineation and prescription file generation.

Post-training discriminative pruning for RBMs

One of the major challenges in the area of artificial neural networks is the identification of a suitable architecture for a specific problem. Choosing an unsuitable topology can exponentially increase the training cost, and even hinder network convergence. On the other hand, recent research indicates that larger or deeper nets can map the problem features into a more appropriate space, and thereby improve the classification process, thus leading to an apparent dichotomy. In this regard, it is interesting to inquire whether independent measures, such as mutual information, could provide a clue to finding the most discriminative neurons in a network. In the present work, we explore this question in the context of Restricted Boltzmann machines, by employing different measures to realize post-training pruning. The neurons which are determined by each measure to be the most discriminative, are combined and a classifier is applied to the ensuing network to determine its usefulness. We find that two measures in particular seem to be good indicators of the most discriminative neurons, producing savings of generally more than 50% of the neurons, while maintaining an acceptable error rate. Further, it is borne out that starting with a larger network architecture and then pruning is more advantageous than using a smaller network to begin with. Finally, a quantitative index is introduced which can provide information on choosing a suitable pruned network.

Pigmented skin lesion segmentation based on sparse texture representations

Among the most dangerous cancers, there is the Melanoma that affects millions of people. As this is a type of malignant pigmented skin lesion and it can be recognized by medical experts, computer-aided diagnostic systems are developed in order to assist dermatologists in clinical routine. One of the more difficult tasks is to find the right segmentation of lesions whose precision is very important to distinguish benign from malignant cases. In this work, we propose a new method based on sparse representation. First, an alternative representation of the image is obtained from the texture information. A sparse non-negative dictionary is computed and every image is projected onto this space. The reconstruction is calculated using only the most active atoms, which allows to obtaining an enhanced version of the texture where the morphological post-processing can effectively extract the lesion area. The experiments were carried out on a publicly available database and performance was evaluated in terms of segmentation error, accuracy, and specificity. Results showed that this first approach performs better than methods reported in the literature on this same data.