Towards automatic annotation of prosodic prominence levels in Austrian German (original) (raw)

10th International Conference on Speech Prosody 2020, 2020

Abstract

The creation of prosodic annotations is one of the most difficult and time-consuming aspects of creating a speech database. Generally, only the speech signal and manually created transcriptions are available in an early resource development stage. This paper presents a tool for annotating prosodic prominence at the word level, using exclusively acoustic features (96 f0-, intensityand durational features). The best performance for separating prominent from non-prominent words in Austrian read speech was reached with a decision tree with the absolute word duration as the only feature. For distinguishing more prominence levels, a good performance was reached with a random forest model, similar to the best inter-annotator agreement. Furthermore, we analyzed in detail the feature ranking of the random forest to give us insights into the relative importance of the features contributing to prominence in Austrian German: Word duration > f0 range, RMS range. The specific findings of this study will mainly be relevant for speech scientists and prosody researchers interested in German. Our methodological approach of analyzing prosodic prominence from a purely acoustic perspective at the word-level will also be interesting for researchers focusing on prosody in other languages.

Anneliese Kelterer hasn't uploaded this paper.

Let Anneliese know you want this paper to be uploaded.

Ask for this paper to be uploaded.