Prosodic cues of non-native accent: a pilot study of Finnish read by Estonians

    E.Meister, [1]

    [1] Laboratory of Phonetics and Speech Technology
    Institute of Cybernetics at Tallinn Technical University

    [2] Department of Phonetics
    University of Helsinki


    Native listeners can easily recognize foreign accent of non-native speakers. Many Estonians can understand and speak Finnish at some level, as our languages are so close. Still, there are several differences in sound system and prosodic structure, which ought to be the main sources of non-native accent. This pilot study concentrates on prosodic cues in speech of Estonian speakers when reading Finnish text.

    Speech material and subjects
    4 native speakers of Estonians (2 male and 2 female) and 2 native speakers of Finnish (1 male and 1 female) reading the same text in Finnish have been recorded. The Estonian speakers have been never studied Finnish, their knowledge of Finnish is obtained mainly by watching Finnish TV-channels during several years. Also personal contacts with Finnish people have improved their language skills. The speech recordings of native Finnish speakers serve as the reference data.

    Scope of the study
    If non-native accent is perceptually recognized, several acoustic features should also manifest it. In current acoustic-phonetic study we are looking for answers to the following questions:

  • Which prosodic cues exhibit more non-nativeness?
  • Is speaker profile language-dependent?
  • Are speakers' native languages detectable from his/her non-native speech?
  • Is non-native accent automatically detectable?

  • Methods
    Speech material of Finnish and Estonian speakers has been segmented in four levels (phoneme, syllable, word, breath group). Duration and fundamental frequency for phonemes and syllables have been measured and statistical analysis of duration ratios as well as F0-patterns has been carried out.

    Preliminary results show systematic differences in duration ratios of stressed and unstressed long/short segments of native and non-native speech. Differences in F0-patterns do exist, as well.

    In the presentation the preliminary findings will be introduced and ideas for further research will be discussed.