Separation of linguistic and other information in speech

A previous reseach project (Paralinguistic variation in speech) has produced detailed knowledge about how the speaker's age, sex, vocal effort, type of phonation, etc. affects the acoustic properties of speech (the frequency of the voice fundamental and the formants, the spectral energy distribution, segment durations, etc.). The results from that project form the basis of an investigation of the importance of these acoustic variables for the perception of the different non-linguistic types of speech quality: the perspectival (direction and distance from the speaker), the organic (the speaker's age and sex) and the expressive (vocal effort, type of phonation, speech rate, etc.). The project is focused on man's capacity of separating the linguistic information from all this additional information, which largely affects the same acoustic variables that convey the linguistic message of an utterance. The project aims also at testing a new theoretical model of man's production, perception, and imitation of speech, which allows to simulate this capacity.

The results of this project will be useful for improving technological applications such as automatic recognition of speech, not only in order to recognize the paralinguistic information, but also in order to recognize the linguistic information without being misled by the paralinguistic variation.

Project director: Hartmut Traunmüller
In collaboration with: Anders Eriksson, Ola Andersson, Diana Krull, Gunilla Thunberg

Financially supported by HSFR, the Swedish Council for Research in the Humanities and Social Sciences, for the years 1999-2001. Div. of Phonetics | Dept. of Linguistics | Stockholm University