Untitled

Funder

Organizational Unit

Publications

Prosodic Classification of Discourse Markers

Publication . Cabarrão, Vera; Moniz, Helena; Ferreira, Jaime; Batista, Fernando; Trancoso, Isabel; Mata, Ana Isabel; Curto, Sérgio

The first contribution of this study is the description of the prosodic behavior of discourse markers present in two speech corpora of European Portuguese (EP) in different domains (university lectures, and map-task dialogues). The second contribution is a multiclass classification to verify, given their prosodic features, which words in both corpora are classified as discourse markers, which are disfluencies, and which correspond to words that are neither markers nor disfluencies (chunks). Our goal is to automatically predict discourse markers and include them in rich transcripts, along with other structural metadata events (e.g., disfluencies and punctuation marks) that are already encompassed in the language models of our in-house speech recognizer. Results show that the automatic classification of discourse markers is better for the lectures corpus (87%) than for the dialogue corpus (84%). Nonetheless, in both corpora, discourse markers are more easily confused with chunks than with disfluencies.

2015Journal article

Open access

Automatic Recognition of Prosodic Patterns in Semantic Verbal Fluency Tests - an Animal Naming Task for Edutainment Applications

Publication . Moniz, Helena; Pompili, Anna; Batista, Fernando; Trancoso, Isabel; Abad, Alberto; Amorim, Cristiana

This paper automatically detects prosodic patterns in the domain of semantic fluency tests. Verbal fluency tests aim at evaluating the spontaneous production of words under constrained conditions. Mostly used for assessing cognitive impairment, they can be used in a plethora of domains, as edutainment applications or games with educational purposes. This work discriminates between list effects, disfluencies, and other linguistic events in an animal naming task. Recordings from 42 Portuguese speakers were automatically recognized and AuToBI was applied in order to detect prosodic patterns, using both European Portuguese and English models. Both models allowed to differentiate list effects from the other events, mostly represented by the tunes: L* H/L(-%) (English models) or L*+H H/L(-%) (Portuguese models). However, English models proved to be more suitable because they rely in substantial more training material.

2015Journal article

Open access

Disfluency Detection Across Domains

Publication . Moniz, Helena; Ferreira, Jaime; Batista, Fernando; Trancoso, Isabel

This paper focuses on disfluency detection across distinct domains using a large set of openSMILE features, derived from the Interspeech 2013 Paralinguistic challenge. Amongst different machine learning methods being applied, SVMs achieved the best performance. Feature selection experiments revealed that the dimensionality of the larger set of features can be further reduced at the cost of a small degradation. Different models trained with one corpus were tested on the other corpus, revealing that models can be quite robust across corpora for this task, despite their distinct nature. We have conducted additional experiments aiming at disfluency prediction in the context of IVR systems, and results reveal that there is no substantial degradation on the performance, encouraging the use of the models in IVR domains.

2015Journal article

Open access

Combining Multiple Approaches to Predict the Degree of Nativeness

Publication . Ribeiro, Eugénio; Ferreira, Jaime; Olcoz, Julia; Abad, Alberto; Moniz, Helena; Batista, Fernando; Trancoso, Isabel

Automatic speaker nativeness assessment has multiple applications, such as second language learning and IVR systems. In this paper we view this as a regression problem, since the available labels are on a continuous scale. Multiple approaches were applied, such as phonotactic models, i-vectors, and goodness of pronunciation, covering both segmental and suprasegmental features. Different phonotactic models were adopted, either trained with the challenge data, or using additional multilingual data from other domains. The obtained values were later combined in multiple ways and fed to a support vector machine regressor. Results on the test set surpass the provided baseline and are in line with the results obtained on the remaining sets. This suggests that our models generalize well to other datasets

2015Journal article

Open access

Speech Features for Discriminating Stress Using Branch and Bound Wrapper Search

Publication . Julião, Mariana; Silva, Jorge; Aguiar, Ana; Moniz, Helena; Batista, Fernando

Stress detection from speech is a less explored field than Automatic Emotion Recognition and it is still not clear which features are better stress discriminants. VOCE aims at doing speech classification as stressed or not-stressed in real-time, using acoustic-prosodic features only. We therefore look for the best discriminating feature subsets from a set of 6285 features – 6125 features extracted with openSMILE toolkit and 160 Teager Energy Operator (TEO) features. We use a mutual information filter and a branch and bound wrapper heuristic with an SVM classifier to perform feature selection. Since many feature sets are selected, we analyse them in terms of chosen features and classifier performance concerning also true positive and false positive rates. The results show that the best feature types for our application case are Audio Spectral, MFCC, PCM and TEO. We reached results as high as 70.36% for generalisation accuracy

2015Journal article

Open access