Emotion recognition by speech signals

By Oh-Wook Kwon, Kwokleung Chan, Jiucang Hao, Te-Won Lee. 2003

418 citations recorded by the source corpus.

For emotion recognition, we selected pitch, log energy, formant, mel-band energies, and mel frequency cepstral co-efﬁcients (MFCCs) as the base features, and added velocity/acceleration of pitch and MFCCs to form feature streams. We extracted statistics used for discriminative classiﬁers, assuming that each stream is a one-dimensional signal. Extracted features were analyzed by using quadratic discriminant analysis (QDA) and support vector machine (SVM). Experimental results showed that pitch and energy were the most important factors. Using two different kinds of databases, we compared emotion recognition performance of various classiﬁers: SVM, linear discriminant analysis (LDA), QDA and hidden Markov model (HMM). With the text-independent SUSAS database, we achieved the best accuracy of 96.3% for stressed/neutral style classiﬁcation and 70.1% for 4-class speaking style clas-siﬁcation using Gaussian SVM, which is superior to the previous results. With the speaker-independent AIBO database, we achieved 42.3% accuracy for 5-class emotion recognition.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Privacy-Preserving Adversarial Representation Learning in ASR: Reality or Illusion? (2019, neutral): “…speech processing algorithms can infer not only the spoken contents from the speech signal, but also the speaker’s identity [5], intention [6–9], gender [10,11], emotional state [12–14], pathological condition [15–17], personality [18,19] and cultural [20,21] attributes to a great extent.”
Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge (2011, methodology): “Principal Component Analysis (PCA) (Jolliﬀe, 2002) and Linear or Heteroscedastic Discriminant Analysis (LDA) (Fukunaga, 1990; Ayadi et al., 2007) are the most common techniques, used, e.g., in (Batliner et al., 2000; Kwon et al., 2003; Lee and Narayanan, 2005;Steidl, 2009).”
Comparison of Gender- and Speaker-adaptive Emotion Recognition (2014, background): “One of the pilot experiments which deals with speech based emotion recognition has been presented by Kwon et al. (2003).”
Databases, features and classifiers for speech emotion recognition: a review (2018, background): “Kwon et al. (2003) explored log energy, formants, F0 information, mel based energy, MFCCs with their velocity and acceleration coefficient for speech emotion recognition.”
Features and classifiers for emotion recognition from speech: a survey from 2000 to 2011 (2012, methodology): “To this end, traditional segmental features like MFCCs or LPCCs are transformed to suprasegmental parameterizations through long-time statistic processing, in order to be concatenated to the prosodic feature vector (Kwon et al. 2003; Vogt and André 2006; Schuller et al. 2005a).”
A comparative study of different machine learning methods for landslide susceptibility assessment: A case study of Uttarakhand area (India) (2016, background): “The principle of BE is to find and remove the features that have the most decreases or least increases of classification accuracy of models (Kwon et al., 2003).”
Artificial intelligence in the creative industries: a review (2020, unknown): “tone, loudness, and tempo) [279], face detection (e.”
Automatic analysis of multimodal group actions in meetings (2005, background): “Recent research in recognizing emotion from speech [33], [34], recognizing interest level from posture [35], recognizing hot-spots (regions of high involvement or emphasis) in meetings [36], [37], [38], and detecting agreement and disagreement in meetings [39], suggests that the automatic recognitio”
Emotion recognition from speech: a review (2012, unknown): “Along with F0 information, log energy, formants, mel based energy, MFCCs with their velocity and acceleration coefficients are explored for emotion classification (Kwon et al. 2003).”
Self-Supervised ECG Representation Learning for Emotion Recognition (2020, background): “These data sources include facial expressions [12], motion and gait [13], speech [14], electrocardiogram (ECG) [15], electroencephalogram (EEG) [6], electrooculogram (EOG) [16], and galvanic skin response (GSR) [17] among others.”