Speech as speech – acoustic modeling in variational linguistics

Speech is typically transcribed on a phonetic level in variational linguistics. Phonetic transcribing is very time-consuming and dependent on experts. Meanwhile, the field of automatic speech recognition has recently taken leaps forward. However, the focus of the field has been in the creation of standardized text, even if the original speech signal contains non-standard features. I aim to use automatic speech recognition to output phonetic transcriptions in this project. I moreover utilize neural acoustic modeling to study language variation and change directly from the speech signal and machine translation methods to standardize existing transcriptions to the international phonetic alphabet (IPA). I use existing interview data of Finnish and Norwegian dialects and of Finnish spoken in Helsinki. The project develops new methods for scientists and the public, and provides new insights into language change. The work will take place at Tampere University.