Emotion recognition from speech using global and local prosodic features

Rao, K. Sreenivasa; Koolagudi, Shashidhar G.; Vempada, Ramu Reddy

doi:10.1007/s10772-012-9172-2

Cited by 150 publications

(49 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…One possible approach is to divide the speech signal into many small intervals, i.e., frames, and construct a local feature vector for each frame. For example, prosodic speech features such as pitch and energy, can be extracted from each interval and are considered as local features [26], [30]. On the other hand, global features such as statistics, can be obtained from the whole speech utterance [22], [23], [16], which typically have a lower dimension than the local ones, leading to less computing time [13], [30].…”

Section: B Feature Extractionmentioning

confidence: 99%

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Mao

Dong

Huang

et al. 2014

IEEE Trans. Multimedia

536

198

View full text Add to dashboard Cite

As an essential way of human emotional behavior understanding, speech emotion recognition (SER) has attracted a great deal of attention in human-centered signal processing. Accuracy in SER heavily depends on finding good affect-related, discriminative features. In this paper, we propose to learn affect-salient features for SER using convolutional neural networks (CNN). The training of CNN involves two stages. In the first stage, unlabeled samples are used to learn local invariant features (LIF) using a variant of sparse auto-encoder (SAE) with reconstruction penalization. In the second step, LIF is used as the input to a feature extractor, salient discriminative feature analysis (SDFA), to learn affect-salient, discriminative features using a novel objective function that encourages feature saliency, orthogonality, and discrimination for SER. Our experimental results on benchmark datasets show that our approach leads to stable and robust recognition performance in complex scenes (e.g., with speaker and language variation, and environment distortion) and outperforms several well-established SER features.Index Terms-Affective-salient discriminative feature analysis, convolutional neural networks, feature learning, speech emotion recognition.

show abstract

Section: B Feature Extractionmentioning

confidence: 99%

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Mao

Dong

Huang

et al. 2014

IEEE Trans. Multimedia

536

198

View full text Add to dashboard Cite

show abstract

“…Possible improvements of the proposed work would be the inclusion of other features, as those related to glottal, spectral and energy parameters, as well as a more detailed description of the temporal dynamics of F0 at sentence and word level [18].…”

Section: Discussionmentioning

confidence: 99%

“…Recently, the relevance of shape, slope and range of F0 contour in emotional speech perception, synthesis and recognition has been described [11,[16][17][18]. Moreover, local prosodic features that are related to the temporal dynamics description of prosody have been found to improve the information carried by global, static prosodic features [18].…”

Section: Introductionmentioning

confidence: 99%

Automatic analysis of speech F0 contour for the characterization of mood changes in bipolar patients

Guidi

Vanello

Bertschy

et al. 2015

Biomedical Signal Processing and Control

View full text Add to dashboard Cite

da inserireBipolar disorders are characterized by a mood swing, ranging from mania to depression. A system that could monitor and eventually predict these changes would be useful to improve therapy and avoid dangerous events. Speech might convey relevant information about subjects' mood and there is a growing interest to study its changes in presence of mood disorders. In this work we present an automatic method to characterize fundamental frequency (F0) dynamics in voiced part of syllables. The method performs a segmentation of voiced sounds from running speech samples and estimates two categories of features. The first category is borrowed from Taylor's Tilt intonational model. However, the meaning of the proposed features is different from the meaning of Taylor's ones since the former are estimated from all voiced segments without performing any analysis of intonation. A second category of features takes into account the speed of change of F0. In this work, the proposed features are first estimated from an emotional speech database. Then, an analysis on speech samples acquired from eleven psychiatric patients experiencing different mood states, and eighteen healthy control subjects is introduced. Subjects had to perform a text reading task and a picture commenting task. The results of the analysis on the emotional speech database indicate that the proposed features can discriminate between high and low arousal emotions. This was verified both at single subject and group level. An intra-subject analysis was performed on bipolar patients and it highlighted significant changes of the features with different mood states, although this was not observed for all the subjects. The directions of the changes estimated for different patients experiencing the same mood swing, were not coherent and were task-dependent. Interestingly, a single-subject analysis performed on healthy controls and on bipolar patients recorded twice with the same mood label, resulted in a very small number of significant differences. In particular a very good specificity was highlighted for the Taylor-inspired features and for a subset of the second category of features, thus strengthening the significance of the results obtained with patients. Even if the number of enrolled patients is small, this work suggests that the proposed features might give a relevant contribution to the demanding research field of speech-based mood classifiers. Moreover, the results here presented indicate that a model of speech changes in bipolar patients might be subject-specific and that a richer characterization of subject status could be necessary to explain the observed variability

show abstract

“…Formants of voice are another family features [16]. Other features such as SNR, HNR, speech rate and speech quality are prosodic features [22,23,24]. MFCC, LPCC and LFPC are frequency features of voice.…”

Section: A Features Of Voicementioning

confidence: 99%

Persian speech emotion recognition

Savargiv

Bastanfard

2015

2015 7th Conference on Information and Knowledge Technology (IKT)

View full text Add to dashboard Cite

Speech emotion recognition is one of the most challenging and the most interesting topics of the voice processing research in recent years. Performance enhancement and computational complexity mitigation are the subject matter of the current study. Current study proposes a speech emotion recognition method by employing HMM-based classifier and minimum number of features in the Persian language. Result illustrate the proposed method is able to recognizing eight emotional states of anger, happy, sadness, neutral, surprise, disgust, fear and boredom up to 79.50% average accuracy. In contrast to previous researches, the proposed method provides 8.72% improvement.

show abstract

Emotion recognition from speech using global and local prosodic features

Cited by 150 publications

References 25 publications

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Learning Salient Features for Speech Emotion <newline/>Recognition Using Convolutional <newline/>Neural Networks

Automatic analysis of speech F0 contour for the characterization of mood changes in bipolar patients

Persian speech emotion recognition

Contact Info

Product

Resources

About