Etude de la performance des modèles acoustiques pour des voix de personnes âgées en vue de l'adaptation des systèmes de RAP (Assessment of the acoustic models performance in the ageing voice case for ASR system adaptation) [in French]

Abstract : Our study concerns the integration of an automatic speech recognition system in a social inclusion product designed for elderly people. Due to voice change with age, speech recognition systems present higher word error rate when speech is uttered by elderly speakers compared to when non-aged voice is considered. To characterise these differences in speech recognition performance, we studied which phonemes lead to the lowest recognition rate in the elderly speakers with respect to the younger ones and we collected a specific corpus to make the adaptation of the acoustic models possible. The results show that some phonemes (such as plosives) are more specifically affected by age than others. Finally, the corpus was used to adapt the ASR to the elderly population which resulted in a 5% decrease of the word error rate.
Document type :
Conference papers
Complete list of metadatas

Cited literature [8 references]  Display  Hide  Download

https://hal.inria.fr/hal-00953510
Contributor : Michel Vacher <>
Submitted on : Monday, March 3, 2014 - 5:58:55 PM
Last modification on : Monday, February 11, 2019 - 4:36:02 PM
Long-term archiving on : Saturday, May 31, 2014 - 10:46:42 AM

File

F12-1089.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00953510, version 1

Collections

Citation

Frédéric Aman, Michel Vacher, Solange Rossato, Remus Dugheanu, François Portet, et al.. Etude de la performance des modèles acoustiques pour des voix de personnes âgées en vue de l'adaptation des systèmes de RAP (Assessment of the acoustic models performance in the ageing voice case for ASR system adaptation) [in French]. Actes de la conférence conjointe JEP-TALN-RECITAL 2012, volume 1: JEP, Jun 2012, Grenoble, France. pp.707--714. ⟨hal-00953510⟩

Share

Metrics

Record views

280

Files downloads

321