A5011739249- Speech Recognition and Synthesis
- Speech and Audio Processing
- Speech and dialogue systems
- Music and Audio Processing
- Phonetics and Phonology Research
- Natural Language Processing Techniques
- Voice and Speech Disorders
- Linguistic Studies and Language Acquisition
- User Authentication and Security Systems
- Advanced Clustering Algorithms Research
- Health and Medical Education
- Language Development and Disorders
Universidad de Zaragoza
2007-2012
de niveau recherche, publiés ou non, émanant des établissements d'enseignement et recherche français étrangers, laboratoires publics privés.
In this paper, we present a framework for unsupervised domain adaptation of PLDA based i-vector speaker recognition systems.Given an existing out-of-domain system, use it to cluster unlabeled in-domain data, and then data adapt the parameters system.We explore two versions agglomerative hierarchical clustering that also study automatic ways determine number clusters in dataset.The proposed techniques are experimentally validated recently introduced challenge.This challenge provides very...
This paper addresses the problem that disabled people face when accessing new systems and technologies are available nowadays. The use of speech technologies, specially helpful for motor handicapped people, becomes unapproachable these also suffer impairments, making gap in society wider them. As a way to include impaired technological today, two lines work have been carried out. On one hand, computer-aided therapy software has developed training children with different disabilities. tool,...
This work presents the results in analysis of acoustic features (formants and three suprasegmental features: tone, intensity duration) vowel production a group 14 young speakers suffering different kinds speech impairments due to physical cognitive disorders. A corpus with unimpaired children's is used determine reference values for these without any kind impairment within same domain impaired speakers; this 57 isolated words. The signal processing extract formant pitch based on Linear...
This article presents a low-latency speaker diarization system (“who is speaking now?”) based on hybrid approach that combines traditional offline spoke when?”) with an online identification system. The fulfills all requirements of the task, i.e. it does not need any a-priori information about input, including no specific models. After initialization phase allows decision current accuracy close to underlying describes approach, evaluates robustness system, and analyzes latency/accuracy...
This paper addresses the problem of speaker segmentation in two-speaker telephone conversations, using an eigenvoice based factor analysis approach. We present a set improvements system. First, we study two methods to compensate for intra-session variability, that is variability during single session. Secondly propose method generate hypotheses combined with given confidence measure, enables selection correct improving overall performance. The proposed are evaluated on NIST Speaker...
There are many applications related to speaker characterization, specially in telephone environments, where large datasets available but not directly useful since there two speakers involved every recording. Even with very accurate diarization systems, we can expect find some recordings low accuracy. The use of these may reduce the accuracy any characterization technology. Therefore, it is highly desirable detect those correctly segmented, order discard or process manually remaining ones...
This paper presents the results of an experience with "Vocal-izaL2", application for Second Language (L2) learning Spanish, in a multilingual environment at Vienna International School (VIS).For experiment, group 6th-graders school practiced during 5 sessions altogether their regular classes.The experiment show on one hand, great motivation power that computer-based L2 tools have pronunciation training young learners, while also resulting useful teachers.On technical aspect, tool and...