A set of tools for the extraction and analysis of prosodic features is being developed. The tools are developed jointly in Higgins and a number of other projects (e.g. GROG and Chill).
Some of the tools described here are implemented in Perl, others are plug-ins for Jonas Beskow's and Kåre Sjölander's Wavesurfer, relying heavily on Snack by Kåre. We're also meaning to utilise Kåre Sjölander's aligner nalign.
At present, this page contains mainly some scattered ideas, some implemented, some not. The page will become more structured as work on the prosody tools developes.
Prosody tools and the related protocols are considered to be pre-alpha versions, except for the Silence plug-in, which is a rough beta.
- Wavesurfer silence detector - A simple silence detector implemented as a Wavesurfer plugin.
Perl filters and compilers
The Perl module ProsodyTools contains, for the time being, the following:
- merging of nalign/Wavesurfer .lab and .ord files to a common data structure
- serialising of the data structure
- modularised rule based mark-up of phoneme sequences
- persistancy (file writing) of rule-based mark-up
- merging of mark-up and merged .lab and .ord files
- serialising of the resulting data structure
- de-serialising of saved data
The functionality described above is used as a preprocessing step for prosodic feature extraction and analysis. The rule based mark-up is fully modularised, so that any pattern matching or other rules may be plugged in with little effort. The code is contained in a Perl module, which may be loaded and used from other Perl programs (or programs that can use a Perl interpreter), but it also comes with a set of scripts that can be ran directly from the command line.
The code is alpha status and not publically available yet.
Copyright © 2002-2004 Jens Edlund, Gabriel Skantze and the members of the Center for Speech Technology