Ensemble Bag-of-Audio-Words Representation Improves Paralinguistic Classification Accuracy

Gosztolya, Gábor; Busa-Fekete, Róbert

doi:10.1109/taslp.2020.3044465

Cited by 2 publications

(3 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…From experience (e.g. [12]) we know that it might be beneficial to use multiple different (utterance-level) feature sets, as these might represent the individual utterances from a different aspect, and improve classification. We decided to opt for late fusion [12]: we trained independent SVM models for the different types of features, and combined the predictions in the second step.…”

Section: Prediction Combinationmentioning

confidence: 99%

“…[12]) we know that it might be beneficial to use multiple different (utterance-level) feature sets, as these might represent the individual utterances from a different aspect, and improve classification. We decided to opt for late fusion [12]: we trained independent SVM models for the different types of features, and combined the predictions in the second step. Following our previous studies, we took the weighted mean of the posterior estimates; the weights were determined on the development set with 0.05 increments.…”

Section: Prediction Combinationmentioning

confidence: 99%

“…mean, standard deviation, percentiles) for specific frame-level attributes. Other general feature sets include Bag-of-Audio-Words (BoAW, [9]) and Fisher vectors (FV, [10]), both of which were applied in several studies [2,11,12,13].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Identifying Conflict Escalation and Primates by Using Ensemble X-Vectors and Fisher Vector Features

et al. 2021

View full text Add to dashboard Cite

Computational paralinguistics is concerned with the automatic identification of non-verbal information in human speech. The Interspeech ComParE challenge features new paralinguistic tasks each year; this time, among others, a cross-corpus conflict escalation task and the identification of primates based solely on audio are the actual problems set. In our entry to ComParE 2021, we utilize x-vectors and Fisher vectors as features. To improve the robustness of the predictions, we also experiment with building an ensemble of classifiers from the x-vectors. Lastly, we exploit the fact that the Escalation Sub-Challenge is a conflict detection task, and incorporate the SSPNet Conflict Corpus in our training workflow. Using these approaches, at the time of writing, we had already surpassed the official Challenge baselines on both tasks, which demonstrates the efficiency of the employed techniques.

show abstract

Section: Prediction Combinationmentioning

confidence: 99%

Section: Prediction Combinationmentioning

confidence: 99%