The History of AI

Hidden Markov models in speech

Era 3 · From the expert boom to the statistical turn (1980–2011) › Statistical learning and probabilistic models

Lineage
Data and learning
Kind
Popularization

Who

Bell Labs, IBM, Dragon Systems

What happened

Lawrence Rabiner of Bell Labs publishes "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition" in the Proceedings of the IEEE, which brings together the method speech had been recognized with since the 1980s.

Why it mattered

Speech recognition becomes a statistical problem learned from recordings. Hidden Markov models dominate the field until deep networks arrive.

The full story

In the 1970s speech recognition compared each spoken word with recorded templates, a method that failed with different speakers and large vocabularies. The alternative came from statistics.

A hidden Markov model describes a sequence of states that cannot be seen directly, like the sounds of a word, each of which produces observations that can be seen, like the fragments of an audio signal. Leonard Baum had worked out the mathematics in the late 1960s. James and Janet Baker at Carnegie Mellon and Frederick Jelinek's group at IBM applied the models to speech in the 1970s, and in the 1980s they replaced templates as the dominant method. The IBM group built Tangora, a voice-activated typewriter with a 20,000-word vocabulary. The Bakers founded Dragon Systems in 1982, which in 1990 released Dragon Dictate, a dictation product for the public.

Rabiner's tutorial clearly explained the method's three problems, how to evaluate, how to decode, and how to train a model, and it became the reference for a whole generation of engineers.

These models remained the basis of speech recognizers for two decades, including the one Siri used at launch. Deep networks began replacing their acoustic components around 2010.

Leads to

Sources

How to cite

“Hidden Markov models in speech.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/hidden-markov-models-speech

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline