Course abstracts - Loakes, Jepson & Stoakes
Course: A practical user's guide to MAUS: Theory, applications and troubleshooting
Instructors: Debbie Loakes (University of Melbourne), Hywel Stoakes (University of Auckland), Katie Jepson (University of Melbourne)
Description: This course is designed to give participants an understanding of the web-based Munich Automatic Segmentation System (WebMAUS), a tool to align speech recordings to text at the segmental level (i.e. with reference to individual speech segments). Aligning speech recordings to text or "forced alignment" is the process of getting an orthographic transcription time-aligned to an audio-file on the basis of what is in the acoustic signal. The course will enable participants to confidently use the WebMAUS system to force align their own data. There are three main components of the course: 1) the theory behind MAUS, 2) examples of what WebMAUS can do, and 3) a hands-on introduction to using WebMAUS. On the first day we will take participants through the theory behind MAUS, the statistical models, and ‘backend’ of WebMAUS, including how to go about training a new MAUS language model. Following this, we will show examples of data from a wide range of languages produced/generated using the WebMAUS system (for example, Aboriginal English, Australian English, Djambarrpuyŋu, Drehu, Nafsan, French, and Wubuy), discussing issues which arose in the process, how to troubleshoot some problems, and where to look for help. On the third and fourth days we will run a hands-on workshop in which participants work through all the steps involved in getting .WAV files to forced-aligned, segmented Praat textgrids following a step-by-step workflow developed by our team. By the end of the course, participants will know how to MAUS their own data, have an understanding of how to combat some issues that can arise, and will have a suite of codes to streamline and automate the WebMAUS process.