[Figure: A phone-based Markov source based on the phonetic subsource of Fig. 9.]
Caption: Fig. 11. A phone-based Markov source based on the phonetic subsource of Fig. 9.
Figure text: ACOUSTIC SUBSOURCE OF PHONE t, ACOUSTIC SUBSOURCE OF PHON…A simple method for finding the most likely path is a dynamic programming scheme [10] called the Viterbi Algorithm [11]. Let $\tauk(s)$ be the most probable path to state $s$ which produces output $y1^k$. Let $Vk(s) = P(\tauk(s))$ denote th…Many shortcuts to reduce the amount of computation and storage are possible and we will briefly mention some of the more useful ones. If logarithms of probabilities are used, no multiplications are necessary and the entire search can be car…
Andrew Viterbi
engineer · 3 mentions across 1 reading
In this course
Viterbi is foundational to sequence modeling and decoding in hidden Markov models, providing an efficient dynamic programming algorithm to find the most probable hidden state sequence given observed outputs. The course readings invoke the Viterbi Algorithm in the context of speech recognition and phonetic analysis, where it solves the practical problem of inferring which sequence of phonetic states most likely produced a given acoustic signal. This algorithmic approach bridges information theory and machine learning, enabling the tractable inference problems that underpin many AI systems processing sequential data like speech and text.
Mentioned in 1 reading