Short Description

Language documentation is a rapidly growing field in linguistics due to its urgency. At least half of the world’s 6000+ languages will vanish during the 21st century. Along with the speech communities themselves, linguists are assiduously working to document these languages through recordings, which they translate to global languages like English. A fundamental challenge for language documentation is the “transcription bottleneck”. Transcribing spoken languages is a rather mechanic but very time-consuming task (with a recording-to-transcription-time-ratio up to 1:60).

Speech technology, as a fast developing field in itself, can deliver the decisive breakthrough for overcoming the transcription bottleneck. While good performing ASR systems are based on neural networks trained for large-scale corpora, language documentation is working on low-resource languages.

All primary data for this project is taken from an ongoing documentation project for the Muyu language in New Guinea. Muyu is an endangered language spoken by around 2.000 people in the rain forest of West New Guinea, Indonesia. The main goal is to adapt a system that delivered promising results with data from a Vietnamese language in order to build an automatic transcription system for Muyu. This transcription system will be an integral part of the ongoing documentation project. This leads to a constantly growing body of primary data and opportunity to increase the algorithm’s performance. A working transcription system can also serve as a good base for future projects at the intersection of language documentation and technology.

Your Tasks

Review of Literature

Building acoustic models with data from well documented languages (German, English, Spanish, etc.)

Performing a recognition task in wav2vec2.0 for Muyu language

Your Profile (recommended)

Speech Communication background (or visiting the lecture ASR (Pernkopf, WS, VO) in parallel)

Good programming skills (e.g., in Python).