Vintsyuk: connected speech from dictionary words
In Kibernetika in 1971 (No. 2, March-April) T. K. Vintsyuk posed the problem of recognising connected speech composed of words from a given dictionary and proposed a solution: the division of the signal into words is chosen in the recognition process itself by dynamic programming, using a new computing scheme with a 'potential-optimal index'. The result of recognition is the word sequence.
Why it matters
Dynamic programming, which in 1968 had aligned single words in time, became a way to search at once for the words and the boundaries between them in continuous speech. This is an editorial assessment.
Only the first two pages of the English translation (Cybernetics, vol. 7, no. 2, pp. 361-372) were read. They say that the article continues the author's earlier work, that of all ways to divide the signal into words the one giving the best recognition result is chosen, and that spectrograms of the same word can be approximately converted into one another by non-linear stretching and contraction along the time axis. What the record does not claim: the dictionary size, accuracy or speed, which are on pages not read; where the author worked, since the open pages print no affiliation; that the system was the world's first, as the Encyclopedia of Modern Ukraine writes.