Flemish accent identification based on formant and duration features

Ghesquiere,; Compernolle, Van

doi:10.1109/icassp.2002.1005848

Cited by 11 publications

(6 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Ghesquiere (Ghesquiére and Van Compernolle, 2002) selected informative normalized formant frequencies and duration to identify Flemish accents in a text dependent mode and an absolute improvement of 20.2% was obtained; a similar experiment was done by Hansen (Hansen et al, 2004), who showed that only using a few discriminative phonemes based on the Fisher criterion could boost the system from 30% to 42% for identifying seven American English dialects also in a text dependent mode. These results imply that not all phonemes are discriminative among accents and the existence of non-discriminative phonemes may heavily degrade the performance.…”

Section: Identifying Languages and Accentsmentioning

confidence: 90%

“…For example, in previous work (Ghesquiére and Van Compernolle, 2002), formants of sonorant phonemes were used as features. Although formants have the advantage of being robust against disturbances, they can not be reliably estimated and they are bad descriptors for non-sonorant phonemes.…”

Section: Final Discussion and Conclusionmentioning

confidence: 99%

“…To avoid overloading, the contributions of the three states of each phoneme are taken together because if one state is selected, the other two are mostly selected too. Next to the cepstra and energy, we also show the most useful formants by the ANOVA selection, although the AID result of the formants is not presented since it is absolute 15% worse than that of the cepstra (Ghesquiére and Van Compernolle, 2002). The darkness corresponds to the number of times a feature is selected.…”

Section: Accent Identification With Anova Rankingmentioning

confidence: 99%

See 2 more Smart Citations

Feature subset selection for improved native accent identification

Duchateau

Martens

et al. 2010

Speech Communication

View full text Add to dashboard Cite

In this paper, we develop methods to identify accents of native speakers. Accent identification differs from other speaker classification tasks because accents may differ in a limited number of phonemes only and moreover the differences can be quite subtle. In this paper, it is shown that in such cases it is essential to select a small subset of discriminative features that can be reliably estimated and at the same time discard non-discriminative and noisy features. For identification purposes a speaker is modeled by a supervector containing the mean values for the features for all phonemes. Initial accent models are obtained as class means from the speaker supervectors. Then feature subset selection is performed by applying either ANOVA (Analysis of Variance), LDA (Linear Discriminant Analysis), SVM-RFE (Support Vector Machine -Recursive Feature Elimination), or their hybrids, resulting in a reduced dimensionality of the speaker vector and more importantly a significantly enhanced recognition performance. We also compare the performance of GMM, LDA and SVM as classifiers on a full or a reduced feature subset. The methods are tested on a Flemish read speech database with speakers classified in 5 regions. The difficulty of the task is confirmed by a human listening experiment. We show that a relative improvement of more than ACCEPTED MANUSCRIPT20% in accent recognition rate can be achieved with feature subset selection irrespective of the choice of classifier. We finally show that the construction of speaker based supervectors significantly enhances results over a reference GMM system that uses the raw feature vectors directly as input, both in text dependent and independent conditions.

show abstract

Section: Identifying Languages and Accentsmentioning

confidence: 90%

Section: Final Discussion and Conclusionmentioning

confidence: 99%

Section: Accent Identification With Anova Rankingmentioning

confidence: 99%

See 1 more Smart Citation

Feature subset selection for improved native accent identification

Duchateau

Martens

et al. 2010

Speech Communication

View full text Add to dashboard Cite

show abstract

“…Considering interaction effects, the first F 1 and second F 2 formants were most contributing features to accent classification followed by F 5 , F 3 and F 4 . Majority of other more previous studies have argued on the significance of formants to be F 2 and F 3 [8,14,15] or F 3 only [16] or F 1 , F 2 and F 3 [17,18]. Generally, the accuracy was less than 45% on individual formant and overall, formants were useful in accent discrimination but insufficient to be used alone.…”

mentioning

confidence: 91%

Statistical formant descriptors with linear predictive coefficients for accent classification

Yusnita

Paulraj

Yaacob

et al. 2013

2013 IEEE 8th Conference on Industrial Electronics and Applications (ICIEA)

View full text Add to dashboard Cite

Accent is a special trait of human speech that can deliver some information about a speaker's background. At the same time it is one of the profound factors that affects the intelligibility and performance of speech recognition systems (ASRs) if not delicately handled. Normally accent recognizer in the preceding stage offers subsystem training or adaptation strategy to improve the ASRs. Formant analysis is one of the effective techniques used to extract accent information in speech. In this paper we propose a novel way of modifying formants using statistical descriptors and fusion with linear predictive coefficients (LPC). As a result, the deviation of scores from the means can be reduced and resulted in better accuracy rate. This work was based on database of accents in Malaysian English that are ethnically diverse in nature. Experimental results showed that the proposed fusion of LPC with statistically derived fmntRRS has achieved an increase of 7.61% in the accuracy rate over using LPC alone in the quest to classify three-accent problem.

show abstract

“…The research of the past decades made a useful link between foreign accent research and speech signal degradation experiments. First, this step is a response to technological demands: automatic speech recognizers seem to profit from correct accent detection (Ghesquiere and Van Compernolle, 2002) and because most of our communication actually takes place in acoustically adverse environments, it is beneficial to know which features suffer least when the speech signal is imperfect. Second, since diverse conditions may affect the non-native users of a language code differently from the native ones, there is an opportunity to pose interesting research questions.…”

Section: Introductionmentioning

confidence: 99%

The strength of foreign accent in Czech English under adverse listening conditions

Volín

Skarnitzl

2010

Speech Communication

View full text Add to dashboard Cite

The study connects two major topics in current speech research: foreign accentedness and speech in adverse conditions. We parallel the research in intelligibility of non-native speech, but instead of linguistic unit recognition we focus on the perception of the foreign accent strength. First, the question of type and degree of perceptual deficiencies occurring along with certain types of signal degradation is tackled. Second, we measure correlations between the accent ratings and certain candidate phenomena that may influence them, e.g., articulation rate, temporal patterning, contrasts in sound pressure levels on selected syllables and F0 variation. The impacts of different types of signal degradation help to estimate the role of segmental/suprasegmental information in assessments of foreignness in Czech English. The full appreciation of the strength of foreign accent is apparently not possible without fine phonetic detail on the segmental level. However, certain suprasegmental features of foreignness are robust enough to manifest at severe levels of signal degradation. Pairwise variability indices of vowel durations and variation in F0 tracks seem to guide the listener even better in the degraded than in the 'clean' speech signal.

show abstract

Flemish accent identification based on formant and duration features

Cited by 11 publications

References 0 publications

Feature subset selection for improved native accent identification

Feature subset selection for improved native accent identification

Statistical formant descriptors with linear predictive coefficients for accent classification

The strength of foreign accent in Czech English under adverse listening conditions

Contact Info

Product

Resources

About