Search

Automatic speech recognition (ASR) has now reached the point where practical applications can be envisaged. However the models that are presently used still have to be enhanced, especially to improve the robustness of recognition in real conditions. Most of present systems are based on stochastic models, especially hidden Markov models (HMMs). In the past few years, a quite large number of projects have been directed toward the development of a new class of models: the connectionist artificial neural networks (ANNs).

If automatic speech recognition systems perform remarquably well,even for large vocabulary or multi-speaker tasks, their performances degrade dramatically in adverse situations,especially in the presence of noiseor distortion. For instance, problems are created by differences which may occur between training and testing conditions (noise level as measured by the signal-to-noise ratio, SNR,distance to the microphone and orientation, etc.), If trainingand testing can be carried in the same difficult conditions, performances turn out to be significantly better than that obtained when training is noise-free.

Connectionist models, also known as neural networks, have been widely studied during the past few years. Applications concern various tasks in the fields of pattern recognition and signal processing, especially automatic speech recognition. This chapter presents the basic properties of these models and the different problems in the area of speech recognition to which they have been applied so far. The classical models of neural networks are also briefly recalled. We then concentrate on a particular model grounded on neuro-biological data, the cortical column. The characteristics of the model are given and we then present the architectures of two systems based on the cortical column for solving two different problems of speech recognition, i.e. acoustic-phonetic decoding and isolated word recognition.

We propose a novel approach to neuron-like models that consists in simulating cortical columns, i.e. associations of elementary neurons having a specific, functionnal activity. This model is built according to a theory which is consistent with neurobiological data. It provides autonomous units, each computing its output as a function of its inputs, modulated by such internal features as learning coefficients.

Artificial Intelligence (AI) has recently advanced to the point that practical applications are now existing in several domains. Most of the results obtained are not due to general problem solving techniques but to the use of specific, domain-dependent knowledge. Formalizing and incorporating specific knowledge into a system makes it possible to reach the level of expertise comparable to that of a human expert in some specialized field. Such knowledge-based and expert systems have been extensively used in various domains like chemistry, medicine, geology, etc. The basic idea in these systems is to clearly distinguish between the knowledge base which usually incorporates rules and meta-rules about the domain of expertise and the control structures which manipulate this knowledge. That ensures great modularity and flexibility and makes it easy to modify and update a system [11].