DDL - UMR 5596
ISH - Bat C
14 avenue Berthelot
69007 Lyon
Tél : 04 72 72 64 12
Fax : 04 72 72 65 90
Contact

Identification

Themes and actions

 You are here : Home /  Research /  Identification /  Themes and actions

Team webmaster : François PELLEGRINO

  Distance between languages and dialects
  Distance between language and non-language
  Inter-speaker distance

Distance between languages and dialects


Language and dialect identification lies at the interface between the information sciences and technologies, linguistics, and the cognitive sciences. Our research is mainly concerned with two complementary aspects:

  • studying between-langage and between-dialect perceptual distance
  • modelling acoustico-phonetic and phonological cues for automatic identification purposes
The perceptual studies have a twofold goal: assessing the relevance of segmental and suprasegmental cues for language or dialect identification, and comparing the cognitive performances of human listeners with linguistic typologies and computer models. The aim here is to gain a better insight into how foreign languages are represented and what cognitive mechanisms are involved. The automatic identification part is concerned with developing models based on these cues (especially the suprasegmental ones) and integrating them into operational systems for automatic identification. Linguistic domains covered:
  • Afro-Asiatic languages (Contact : M. Barkat Defradas ; N. Louali)
  • European languages (Contact : F. Pellegrino)
  • Arabic dialects (Contact : M. Barkat Defradas ; R. Hamdi)
  • English dialects (Contact : E. Ferragne)

Structure and Dynamics of phonetic-phonological systems
 Egidio MARSICO , François PELLEGRINO
Phonetics, Phonology, Complex Adaptive Systems, Features, Databases
Read more...

Automatic identification
 François PELLEGRINO
automatic typology; language identification; Afro-asiatic languages; European languages; Arabic dialects; English dialects, prosodic modelling; rhythmic modelling
Read more...

Comparative Typology and corpus
 Mélissa BARKAT-DEFRADAS , Naïma LOUALI , François PELLEGRINO
comparative typology; afro-asiatic languages; european languages; arabic dialects; english dialects
Read more...

Distance between language and non-language


Human beings possess the ability to understand the meaning of utterances produced under a wide range of conditions that can be subsumed under two categories:

  • production (between- and within-speaker variability)
  • environment (spatial distance between speaker and hearer, background noise, telephone speech vs. direct conversation, etc.)
This ability relies on a succession of processing stages (from the peripheral auditory to the pragmatic and semantic processing) that are interconnected through bottom-up and top-down mechanisms. Reverse speech is a particularly drastic type of degraded speech signal; yet the linguistic content remains accessible under certain conditions. The experiments we are conducting are intended both to assess this intelligibility through perceptual experiments and to investigate the cerebral correlates of the cognitive processing involved. The study of these mechanisms will lead us to better comprehend how language is processed and to evaluate whether this processing is specific to speech sound and to what extent it applies to non-speech sounds. We will then be able to consider the potential bias introduced in experiments on functional brain imaging by using reverse speech as a control condition.

Evaluating the intelligibility of reverse speech
 Mélissa BARKAT-DEFRADAS , Fanny MEUNIER
Intelligibility; perception; reverse speech; degraded speech; cognitive reconstruction; perceptual experiments
Read more...

Searching for the cerebral correlates of reverse-speech processing
 Mélissa BARKAT-DEFRADAS , Fanny MEUNIER
Intelligibility; perception; reverse speech; degraded speech; cognitive reconstruction; perceptual experiments; Event-Related Potentials; ERP
Read more...

Inter-speaker distance


It is possible to extract several acoustic parameters from recordings of a given person's voice. These parameters are then used to build a model for this speaker and to compute a distance metric between various speakers. This distance is defined mathematically or empirically, and it does not necessarily correspond to perceptual distance. In the field of speaker identification, for instance, it should be noted that recordings of an unknown voice will by no means allow the researcher to infer the speaker's identity with absolute certainty. Although these techniques are highly controversial – especially when employed in a forensic context – they can be used for other purposes such as content-based document indexing, or as an access control for automatic client identification procedures.


Automatic audio document indexing
 Ivan MAGRIN-CHAGNOLLEAU
Audio indexing; speaker tracking; speaker segmentation; statistical modelling
Read more...

Automatic speaker identification
 Ivan MAGRIN-CHAGNOLLEAU
Automatic speaker recognition; speaker verification; speaker tracking; statistical modelling
Read more...

ASLAN -  Université de Lyon -  CNRS -  Université Lumière Lyon 2 -  MSH-LSE -  IXXI -  DDL :  Contact |  Terms of use |