The long-term (since 1988) collaborative efforts of Nelson Morgan and Hervé Bourlard in developing neural network technology (NN) for speech processing laid the foundation for the revolution in speech recognition that has led to the modern dominant approaches that, among many applications, make speech-to-text and speech interactions with smart devices possible. Long before NNs became globally fashionable, they laid down key links between NN, Bayesian statistics, and linear algebra and set out the key architectural ideas of integrating the sequential pattern matching power of a Hidden Markov Model (HMM) with the phonetic discrimination of a NN for automatic speech recognition (ASR). While their methods were originally considered very computationally expensive, the duo continued improving the use of NNs in ASR until advances in computational power allowed a breakthrough using their original HMM/NN architecture. They also led the development of techniques for transcribing in speaker-independent, multispeaker environments, as well as for recognizing speech in noise.
An IEEE Fellow, Bourlard is director of the Idiap Research Institute, Martigny, and full Professor at the EPFL, Switzerland.