1,034,212 research outputs found
DISCOVERING UNDERLYING TONAL REPRESENTATIONS BY COMPUTATIONAL MODELING: A CASE STUDY OF THAI
No full textIn the present study we test a computational method for investigating underlying tonal representations. The representation explored is in the form of simple linear functions as ideal pitch targets, with which close-to-natural F0 contours can be computationally generated. The estimation of the pitch targets is done with PENTAtrainer2, a hypothesisdriven prosody-modeling tool that combines functional annotation, quantitative Target Approximation and global stochastic optimization. In this study we applied PENTAtrainer2 in an investigation of Thai tones. We applied PENTAtrainer2 on a functionally annotated multi-speaker Thai corpus. The pitch targets learned from the corpus showed clear separation between tonal categories, and the F0 contours synthesized with these targets showed close resemblance to those of natural speech of different speakers, whether or not a particular speaker’s data were used in the training. The results demonstrate that it is possible to establish highly economical tonal representations (three parameters per target per tone) that are both fully contrastive and capable of capturing fine phonetic details of Thai tones. Also demonstrated by the study are the effectiveness of PENTAtrainer2 as a prosody research tool, and the potential of computational modeling in general as a new means of basic research in linguistic science. SUBJECT KEYWORD
Acoustic-to-Articulatory Inversion of a Three-dimensional Theoretical Vocal Tract Model Using Deep Learning-based Model
No full textCorrigendum: Economy of Effort or Maximum Rate of Information? Exploring Basic Principles of Articulatory Dynamics
No full textThe Design and Implementation of an Image Processing Framework with a Graphical Programming Interface for Low-Ended FPGAs
No full text
- …
