Multiple windowed spectral features for emotion recognition

Abstract

MFCC (Mel Frequency Cepstral Coefficients) and PLP (Perceptual linear prediction coefficients)or RASTA-PLP have demonstrated good results whether when they are used in combination with prosodic features as suprasegmental (long-term) information or when used stand-alone as segmental (short-time) information. MFCC and PLP feature parameterization aims to represent the speech parameters in a way similar to how sound is perceived by humans. However, MFCC and PLP are usually computed from a Hamming-windowed periodogram spectrum estimate that is characterized by large variance. In this paper we study the effect of averaging spectral estimates obtained using a set of orthogonal tapers (windows) on emotion recognition performance. The multitaper MFCC and PLP are examined separately as short-time information vectors modeled using Gaussian mixture models (GMMs). When tested on the FAU AIBO spontaneous emotion corpus, a relative improvement ranging from 2.2% to 3.9% for both MFCC and PLP systems is achieved by multiple windowed spectral features compared to single windowed ones.

Emotion recognition from spontaneous speech using Hidden Markov models with deep belief networks

“…With speaker-specific z-normalization, we obtained a UAR of 46.36%. Respectively, these results are significantly better than 44.0% and 44.8% UAR, the current state of the art without [3] and with [4] speaker normalization (one-tailed binomial test, p ≈ 0.002).…”

“…Hassan et al [8] achieved a 42.7% UAR by applying importance weights within a SVM to compensate for differences between training and testing conditions. Attabi et al [3] used GMM to model multiple windowed spectrum estimates of Perceptual Linear Prediction (PLP) coefficients, resulting in a 44.0% UAR. The best known result, 44.8% UAR, was achieved with a two-pass system in which a high-level SVM classified each test utterance using ranking scores obtained from five low-level SVMs, one for each emotion [4].…”

Multi-taper spectral features for emotion recognition from speech

“…Ia k (m ) k� 2 (15) viii) Spectral Slope SSL(m): It is a measure of voice quality found using linear regression given by Eq. (16), and it represents the amount of decrease of spectral amplitude based on human perception.…”

“…Multi-tapers have been widely used recently for speaker recognition and verification purposes [12][13][14]. In [15], the authors applied multi-tapers for emotion recognition purpose but using only MFCC and perceptual linear prediction (PLP) features. In this paper, various spectral features are used from both conventional and multi-taper spectral estimates to recognize speech emotions.…”

Multitaper MFCC Features for Acoustic Stress Recognition from Speech

“…The multitaper approach have been used in several domains including geophysical applications [11], speaker verification [12], [13] and emotion recognition [14], [15] and it has been shown to improve the performance and robustness of different systems. However, this method has not been used in stress speech recognition applications.…”