Automatic Speech Recognition – A Brief History of the Technology Development

Finished · ASR

Summary

Very broad overview paper of where ASR models have come since the early 20th century up through the start of the 21st century (2005). The main successes in the field were built around Hidden Markov Models (HMMs), which seem to be ''doubly'' stochastic models based on using observed acoustic signals to predict phones and words, and second, spectral representation models. A huge consensus: more data always improves word accuracy rate.