Two papers from prosodylab have been accepted at Interspeech:
Wagner, Michael, Iturralde Zurita, Alvaro, and Zhang, Sijjia (in press). Parsing speech for grouping and prominence, and the typology of rhythm. Proceedings of Insterspeech in Brno, Czechia. [paper]
You can try out a version of the experiment yourself at the new prosodylab field station.
Abstract:
Humans appear to be wired to perceive acoustic events rhythmically. English speakers, for example, tend to perceive alternating short and long sounds as a series of binary groups with a final beat (iambs), and alternating soft and loud sounds as a series of trochees. This generalization, often called the ‘Iambic-trochaic Law’ (ITL), although viewed as an auditory universal by some, has been argued to be shaped by language experience. Earlier work on the ITL had a crucial limitation, in that it did not tease apart the percepts of grouping and prominence, which the notions of iamb and trochee inherently confound. We explore how intensity and duration relate to percepts of prominence and grouping in six languages (English, French, German, Japanese, Mandarin, and Spanish). The results show that the ITL is not universal, and that cue interpretation is shaped by language experience. However, there are also invariances: Duration appears relatively robust across languages as a cue to prominence (longer syllables are perceived as stressed), and intensity for grouping (louder syllables are perceived as initial). The results show the beginnings of a rhythmic typology based on how the dimensions of grouping and prominence are cued.
A paper about the *prosoBeast* annotation tool:
Gerazov Branislav and Michael Wagner (in press). ProsoBeast Prosody Annotation Tool. Proceedings of Insterspeech in Brno, Czechia. ArXiv e-prints. [paper][git]
The labelling of speech corpora is a laborious and time-consuming process. The ProsoBeast Annotation Tool seeks to ease and accelerate this process by providing an interactive 2D representation of the prosodic landscape of the data, in which contours are distributed based on their similarity. This interactive map allows the user to inspect and label the utterances. The tool integrates several state-of-the-art methods for dimensionality reduction and feature embedding, including variational autoencoders. The user can use these to find a good representation for their data. In addition, as most of these methods are stochastic, each can be used to generate an unlimited number of different prosodic maps. The web app then allows the user to seamlessly switch between these alternative representations in the annotation process. Experiments with a sample prosodically rich dataset have shown that the tool manages to find good representations of varied data and is helpful both for annotation and label correction. The tool is released as free software for use by the community.