Intelligibility of consonants in CVC utterances produced by dyadic rule synthesis
article
The intelligibility of initial and final consonants in monosyllabic CVC utterances was measured for rule-synthesized speech (dyadic concatenation). In order to evaluate the intelligibility results, two additional speech conditions were tested PCM-coded speech (12 bits, 10 kHz sampling rate), and LPC-resynthesized speech (12 coefficients). The stimulus set consisted of all possible initial and final consonants appearing in nine different CVC contexts, namely with three vowels /i, u, /, and with three nonneighboring consonants /p, t, k/. The subjects had to identify the initial or final consonant in each word and could choose any single consonant as a response. The overall percentage correct scores, averaged over 33 listeners, for the initial and final consonants, were 93% and 92.8% for PCM-coded speech, 86.7% and 85.4% for LPC-synthesis, and 58.2% and 73.5% for rule-synthesized speech. Apart from these scores also the actual confusions are discussed in some detail. More important than the scores achieved for this particular, meanwhile improved, version of the rule-synthesis system, was the insight gained in how to evaluate and improve such systems.
Topics
TNO Identifier
5796
Source
Speech Communication, 2(1), pp. 3-13.
Pages
3-13
Files
To receive the publication files, please send an e-mail request to TNO Repository.