Half-Syllabic Units For Speech Processing – an Automatic Segmentation

Authors

  • Mamoru Nakatsui

Abstract

The half-syllabic units proposed here are units each of which has segment boundaries at steady portions and preserves a transition between two phonetic units. Segment boundaries are basically determined by the minima (valleys) of gross spectral variation measure. The spectral variation measure is defined as the root-mean-square value of the slopes of the weighted regression lines calculated from LPC cepstrum parameters over several frames. The maxima (peaks) of the measure will serve as the reference points for further processing, In speech synthesis by rule, it is primarily important to select synthetic units that have reasonably small size of inventory to represent spoken utterances and, at the same time, are easily concatenated. In speech analysis synthesis system at very low-bit-rates such as phonetic vocoding, the units must, further, be automatically segmented and be suitable for interpreting into or matching with the reference units. These requirements on segmentation and matching or labelling are expected to be satisfied for speech recognition system in many cases and for providing useful tools for automatic generation of the inventory of concatenative units.

Additional Files

Published

2022-12-03

How to Cite

1.
Nakatsui M. Half-Syllabic Units For Speech Processing – an Automatic Segmentation. Canadian Acoustics [Internet]. 2022 Dec. 3 [cited 2024 Dec. 3];14(3 bis):51-2. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/3518

Issue

Section

Proceedings of the Acoustics Week in Canada