Loading timeline…
19751970s
Hidden Markov Models
Statistical models of sequences with unobservable states — the backbone of speech recognition for three decades.
Why It Was Important
HMMs gave speech recognition its first rigorous probabilistic foundation, replacing hand-built phonetic rules with models trained on data. Adopted at IBM and CMU in the 1970s, they dominated speech and later bioinformatics until deep learning displaced them around 2012, and they were the clearest early proof that statistical learning could beat expert knowledge on a hard perception task.
Who Invented It
Leonard Baum, Frederick Jelinek, James Baker
Baum developed the underlying mathematics; Jelinek at IBM and Baker at CMU applied it to continuous speech.
Applications
- Speech Recognition
- Part-of-Speech Tagging
- Gene Sequence Analysis
- Handwriting Recognition
Key Papers
- A Maximization Technique Occurring in the Statistical Analysis of Probabilistic Functions of Markov Chains
Leonard E. Baum et al. · The Annals of Mathematical Statistics · 1970