Connectionist and hybrid models for automatic speech recognition

From MaRDI portal





Most of the present models in Automatic Speech Recognition (ASR) are based on stochastic models, especially Hidden Markov Models (HMMs). In the latest years, a quite large number of projects have been directed towards the development of a new class of models: the connectionist Artificial Neural Networks (ANNs). Besides the fact that ANNs represent crude models of the human cortex, there are also other good reasons in favour of ANNs in ASR: (a) ANNs can learn from examples, a feature particularly important to ASR. (b) ANNs, unlike HMMs, do not require strong assumptions about the statistical properties of the input data. (c) The ANN formalism intuitively acceptable, in general. (d) ANNs are capable of producing highly nonlinear functions of the inputs. (e) ANNs provide parallel and regular structures which make them particularly suitable for high performance hardware implementations. The paper proposes an overview of ANNs and their various uses in the field of speech signal analysis. ASR, and speaker verification. Different models working in ASR are presented and compared: multi-layer perceptrons, self-organizing maps, recurrent ANNs, etc. The dynamic aspects of speech are then discussed. Hybrid models combining ANN with stochastic models, especially with HMMs, are finally considered. Future research perspectives conclude the paper. For the entire collection see [Zbl 0927.00044].











This page was built for publication: Connectionist and hybrid models for automatic speech recognition

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4942716)