Ready.
In 1986, the M.S. thesis, "Prosodically Guided Syllable Based Speaker Independent Isolated Turkish Word Recognizer" was submitted to a Turkish university, electronics-engineering dept. The student who submitted that, was the chair of the computer-engineering, as of 2005. Any value in the works of that "would-be-success?"
That "engineer" has taken a grammar-book structuralism as a base to abridge/purge the (Turkish) language. Limiting the word-list that is allowable, only to the words with "wovel agreements/harmony." That is too simple.
Preferrably, engineers were supposed to craft their gadgets, not the human-languages.
A compounding gross strategy is basing all logic on matching the first syllable. If that portion is the piece with the error, the rest is not likely to be found, at all. That "harmony/agreement" utopia, even if it were true, could start from any syllable. A test, variously-pivoting, to match starting at any syllable (that is, not in strict sequential order), and average their results, would be the sensible.
His word-list (p.66) is with only nine command words (bashla, sirala, yeniden, girdi, getir, chikti, sakla, birleshtir, chevir), other than the ten digits (zero to nine). Too few, and quite non-similar words, simplifying a classification. Too simple. Not even to mention that, the "setting of thresholds based after the performance of the system" sounds as a grossly unwise strategy, if the system were meant to really be "speaker-independent." Four people, in all, were both for system-tuning, and for testing. That gadget was tuned to their characteristics!
If the word will be isolated, then simply with consonant-sound classifying, we have the string for running the good old soundex or alternative (Nov.1987) search algoritm to look up a list of words. To submit a 90-page M.S. thesis, is only a gross illusion if it does not improve beyond that.
I consider programming what I have suggested -- for example, for frozen@mid80.
Talking as input, is probably good for young children, and other illiterate people.
To derive the consonant-string from the sound waveform does not sound so hard -- with the invariant features as identified by Cole & Scott (1974, as referred by Margaret W. Matlin "Sensation & Perception, 2nd ed." p.330).
After the few, for any position, are listed, I may trivially test for all of the probable cases. For example, if the first letter is t, the next is either p or b, and the next is b or p, I would test four cases, 1*2*2. Trivial. (In Arabic/Turkish, there is tabib, that is, doctor, to fit that.)
I had criticized also that other, 1994 (Ivy League) paper[/PhD], that was not good quality. That paper and the M.S. thesis discussed, have the link that their authors both got their B.S., that is, their "engineer" title in the same (Turkish) electronics dept, in consecutive years. (The word "success" is magic? Blurred?)