Modeling speech analysis workflows in VisTrails

Reproducibility of results is a fundamental component of conducting comprehensive research. When experimenting with data-analysis pipelines, time spent managing low-level processes and juggling inconsistencies between tools is a barrier to effective research. Rather than exclude persons who might not have a strong technical/programming background to work around these limits, Utilising a "higher level" workflow environment could be of great benefit in certain domains - for this project, that domain is acoustic phonetics. This project is an exploration into the effective use of the workflow sofware "VisTrails" for building acoustic phonetics workflows. VisTrails is a python-based application where workflows can be visually constructed using a "drag and drop" interface of logic modules and linked input/output ports. Users can test workflows under a variety of operational parameters without the need to modify low-level source-code. Furthermore, users can create custom VisTrails packages tailored for particular domains. These can be bundles and shared between colleagues for easily repeatable and extensible experiments. In this project, the core design elements of workflows are investigated within the context of acoustic phonetics, and a target analysis task (classifying phones using formant trajectory data) is implemented at various levels of module granularity. The benefits/drawbacks of each approach are investigated, and assessments are made on the practicality of working with VisTrails workflows, and their component modules, at these levels of abstraction.