Mikio Katagiri

Mikio Katagiri is a Japanese computer scientist and engineer known for his contributions to the field of speech recognition and spoken language processing. He is particularly recognized for his work on discriminative training methods for hidden Markov models (HMMs), which significantly improved the accuracy of speech recognition systems.

Katagiri's research focused on developing algorithms that directly optimize the parameters of HMMs to minimize the error rate on a specific training dataset. This approach contrasted with earlier methods that relied on maximum likelihood estimation, which could lead to suboptimal performance. His work explored various discriminative training criteria, such as maximum mutual information (MMI) and minimum classification error (MCE).

His research has been influential in the development of robust and accurate speech recognition systems, and his publications are widely cited in the speech processing community. He has held academic positions and has also been involved in industrial research and development efforts related to speech technology. Beyond specific algorithms, Katagiri's work contributed to a broader understanding of the principles and techniques involved in building effective speech recognition systems.