Jonathan Harrington
Andrew McVeigh
Speech Hear. and Language Res. Ctr., Macquarie Univ., 2109 Sydney, Australia
The mu+ system has been developed in response to the increasing availability of phonetically and prosodically segmented speech databases. Using an on-line dictionary and pronunciation rules, mu+ maps waves derived label files containing phonetic segments onto phonemes that are in turn parsed into words. Foot and syllable structure are built automatically between word and phoneme levels and a grid is provided to enter (manually) intonational transcriptions. The different kinds of labels can be identified both hierarchically and sequentially. A typical search and display pattern would be: ``display as ellipses in the formant plane F1 and F2 at the vowel target of all accented front vowels that follow syllable-initial oral stops in trisyllabic content words that are phrase-final and utterance-medial.'' Analogous search patterns and displays can also be made for segmented speech production (movement, EMG) data and any combination of segments can be D/A converted. The mu+ system has applications not only in speech recognition and text-to-speech, but also in many classic speech research areas such as coarticulation, vowel undershoot, and speech timing.