Ioana Vasilescu

Fear-type emotion recognition for future audio-based surveillance systems

This paper addresses the issue of automatic emotion recognition in speech. We focus on a type of emotional manifestation which has been rarely studied in speech processing: fear-type emotions occurring during abnormal situations (here, unplanned events where human life is threatened). This study is dedicated to a new application in emotion recognition - public safety. The starting point of this work is the definition and the collection of data illustrating extreme emotional manifestations in threatening situations. For this purpose we develop the SAFE corpus (situation analysis in a fictional and emotional corpus) based on fiction movies. It consists of 7h of recordings organized into 400 audiovisual sequences. The corpus contains recordings of both normal and abnormal situations and provides a large scope of contexts and therefore a large scope of emotional manifestations. In this way, not only it addresses the issue of the lack of corpora illustrating strong emotions, but also it forms an interesting support to study a high variety of emotional manifestations. We define a task-dependent annotation strategy which has the particularity to describe simultaneously the emotion and the situation evolution in context. The emotion recognition system is based on these data and must handle a large scope of unknown speakers and situations in noisy sound environments. It consists of a fear vs. neutral classification. The novelty of our approach relies on dissociated acoustic models of the voiced and unvoiced contents of speech. The two are then merged at the decision step of the classification system. The results are quite promising given the complexity and the diversity of the data: the error rate is about 30%.

De la construction du corpus émotionnel au système de détection. Le point de vue applicatif de la surveillance dans les lieux publics

The study presented in this paper deals with the modelling of extreme emotions occurring in abnormal situations. The aimed application is civil safety and surveillance in the public places in particular. A corpus of fiction (SAFE Corpus) is selected illustrating rich and varied contexts with the presence of extreme emotions, mainly fear. An annotation strategy adapted to the application is then developed, with both generic and specific descriptors. Finally, a detection system of fear emotions based on acoustic cues is implemented to carry out an evaluation. On the one hand the system is robust to context changes. On the other hand, the influence of multimodal annotation is minor. Results obtained with the various protocols are similar: fear is recognized with 67% of success.

Marginal Contrast Among Romanian Vowels: Evidence from ASR and Functional Load.

This work quantifies the phonological contrast between the Romanian central vowels [2] and [1], which are considered separate phonemes, although they are historical allophones with few minimal pairs. We consider the vowels’ functional load within the Romanian inventory and the usefulness of the contrast for automatic speech recognition (ASR). Using a 7 hour corpus of automatically aligned broadcast speech, the relative frequencies of vowels are compared across phonological contexts. Results indicate a near complementary distribution of [2] and [1]: the contrast scores lowest of all pairwise comparisons on measures of functional load, and shows the highest Kullback-Leibler divergence, suggesting that few lexical distinctions depend on the contrast. Thereafter, forced alignment is performed using an existing ASR system. The system selects among [1], [2], ∅ for lexical /1/, testing for its reduction in continuous speech. The same data is transcribed using the ASR system where [2]/[1] are merged, testing the hypothesis that loss of a marginal contrast has little impact on ASR error rates. Both results are consistent with functional load calculations, indicating that the /2/ /1/ contrast is lexically and phonetically weak. These results show how automatic transcription tools can help test phonological predictions using continuous speech.

Large scale data based linguistic investigations using speech technology tools: The case of Romanian

This paper provides a summary of previous efforts made to build an ASR system for Romanian. Thereafter, the data developed within the ASR framework are used to conduct linguistic studies. A first study is dedicated to morpho-phonetic processes in Romanian such as the deletion of masculine definite article -l and the realization of the word final palatalized consonants as plural marker in nouns and person marker in verb conjugation. Data shows that the two phenomena are variable in continuous speech and depend on the degree of spontaneity of the corpus. The second study is dedicated to Romanian vowels acoustic properties. This study takes into account a 7 hours corpus used as development and evaluation data to build the ASR system. Data confirm a seven-vowel system. They also highlight an acoustic proximity and a complementary distribution of the non low central vowels [] and [Λ]. The current findings support previous hypotheses built from laboratory data investigations and encourage further explorations on large scale data.