Speech discrimination by dynamic programming
In January 1968 the Ukrainian journal Kibernetyka published T. K. Vintsyuk's "Speech discrimination by dynamic programming", volume 4, issue 1, pages 81-88; the English translation appeared as Cybernetics volume 4, pages 52-57. It applies dynamic programming to aligning two utterances in time, together with algorithms for recognising connected words.
Why it matters
The central obstacle to recognition - that the same word is never spoken twice to the same length - was removed here by computation rather than by circuitry. The atlas dated this approach to 1978; in fact it was published ten years earlier and remained unknown in the West until the early 1980s, by which time others had proposed the same methods.
The full text is behind the publisher's paywall, so the record rests on the publication page - title, volume, pages and date - and on one 2010 review, which supplies the single claim the primary source does not: that what is meant is non-linear time alignment, and that the work was not known in the West. What the record does not claim: the form of the recurrence, any measured accuracy, any vocabulary, or the institution where the author worked. No accessible source names Vintsyuk's affiliation, so none is given here, only that the journal is Ukrainian, as the publisher of the English translation describes it. The research report behind this batch asserted that Hermann Ney's 1984 paper explicitly acknowledged his algorithm as identical to Vintsyuk's conception; that could not be checked, because the paper is not open access. The reference list visible on the free part of the page cites Kyiv work of 1965.