ISCAN: A system for integrated phonetic analyses across speech corpora


Speech corpora of many languages, styles, and formats exist in the world, representing significant potential for the phonetic sciences. However in practice there are significant practical and methodological barriers to conducting the “same study” across corpora, including necessary technical skills and non-comparability of results using non-standardized measures. We introduce an open-source software system for Integrated Speech Corpus ANalysis (ISCAN), which enables automated acoustic phonetic analysis across spoken corpora of diverse formats and sizes. A web-browser-based GUI and Python package allow for different user backgrounds. The system is a major update of core functionality for fully-automated speech corpus analysis (importing, enriching, querying) from a previous version, to meet new goals - different user configurations, working with restricted datasets, and interacting with data (visualization and correction). The system’s flexibility for different projects is shown in two use cases - large-scale automatic segmental analysis of spontaneous speech across English dialects, and smallerscale semi-automatic prosodic analysis.

In Proceedings of the 19th Congress of Phonetic Sciences Melbourne
Arlie Coles
Arlie Coles
Research Scientist

At the intersection of NLP, deep learning, and clean code.