David Mercier

Audio-Visual Speech Recognition One Pass Learning with Spiking Neurons

We present a new application in the field of impulse neurons: audio-visual speech recognition. The features extracted from the audio (cepstral coefficients) and the video (height, width of the mouth, percentage of black and white pixels in the mouth) are sufficiently simple to consider a real time integration of the complete system. A generic preprocessing makes it possible to convert these features into an impulse sequence treated by the neural network which carries out the classification. The training is done in one pass: the user pronounces once all the words of the dictionary. The tests on the European M2VTS Data Base shows the interest of such a system in audio-visual speech recognition. In the presence of noise in particular, the audio-visual recognition is much better than the recognition based on the audio modality only.

Extracting relevant features to explain electricity price variations

This paper proposes to explain the variations of energy price, namely the electricity on the German market. Such price variations are described by a set of characteristics which are not totally relevant to explain the variations. We first propose to find explanations by using visual tools in order to draw some preliminary conclusions. Analysing such kind of data is usually done thanks to visual comparison by plotting the curves chronologically. In a second time, we propose to build a statistical model from data. The aim of such approach is to detail the characteristic that get involved in the solution, so that we can automatically extract the most pertinent characteristics. We apply this approach on a set of historical data (2007–2010). Obtained results show that methodology is very interesting, since the conclusion from the statistical modelling enforce the visual analysis and also add details about the explanation.

A Generic Pretreatment for Spiking Neuron Application on Lipreading with STANN (Spatio-Temporal Artificial Neural Networks)

Spiking neurons treat sequences of impulses. However the signals to which we have access in the majority of the applications evolve generally continuously with time and are not of impulse nature. If one wants to use spiking neurons, a pretreatment should then be found adapted to the application to convert the raw signals into sequences of impulses. We propose here a simple generic pretreatment which carries out this conversion. We illustrate then this proposal within the framework of the lipreading by STANN (Spatio-Temporal Artificial Neural Networks) and show that this pretreatment is simpler and more effective than that which had been used in [1] for this same application.

Using Kernel Basis with Relevance Vector Machine for Feature Selection

This paper presents an application of multiple kernels like Kernel Basis to the Relevance Vector Machine algorithm. The framework of kernel machines has been a source of many works concerning the merge of various kernels to build the solution. Within these approaches, Kernel Basis is able to combine both local and global kernels. The interest of such approach resides in the ability to deal with a large kind of tasks in the field of model selection, for example the feature selection. We propose here an application of RVM-KB to a feature selection problem, for which all data are decomposed into a set of kernels so that all points of the learning set correspond to a single feature of one data. The final result is the selection of the main features through the relevance vectors selection.