Acoustic Features

Learning English pronunciation necessitates a deep understanding of the prosodic features such as intonation. The present study reports the effective use of BetterAccent Tutor and Praat programs for the learning of English intonation with an experimental and a control group that involves nine students during a third semester phonetics course, at the Department of English, University of Guelma, Algeria. The research aims at investigating the usability of audiovisual feedback paradigm to learn prosody by Algerian students in order to enhance their English intonation renditions. The experimental method enabled generating, measuring and analyzing intonation contours, pitch and intensity of English wh-questions and tag-questions. The results revealed that the audio-visualization technology support helped the students to identify the errors and to understand the functionality of acoustic features they produce regarding intonation patterns.

Music genre is an important feature to identify a musical work. Thus, it is the most used label to organize musical datasets. However, this label is not always available and its identification is not a simple and direct task. Hence, in literature we can find many music genre classification (MGC) methods, with a variety of features and machine learning algorithms (MLA). In this paper, we propose an MGC system by using two levels of hierarchical mining, GLCM (gray level co-occurrence matrix) networks generated from the mel-spectrogram and a multi-hybrid feature strategy. Three types of complex networks were generated: GLCM network Gg, Superpixels network Gs, and GLCM network of each node of Gs (Ggsi network). The multi-hybrid features are formed by textural and topological measures of complex networks and acoustic measures. In the classification step, we used three datasets: GTZAN, Homburg, and ISMIR; two MLAs belonging to the classifier ensemble approach, and (10)-fold cross-validation repeated 100 times. Several experiments were performed using feature combinations of macro-mining (global features of Gg and Gs) and micro-mining (global features of Ggsi). For GTZAN, we performed a detailed analysis of individual class performance and calculated our new ranking logarithmic score (RLS) applied to the F1-score. For all datasets, the RLS and accuracy values were compared with several state-of-the-art methods. The accuracy obtained using micro-mining was >90%, which reveals a satisfactory result.

In the era of data explosion, speech emotion plays crucial commercial significance. Emotion recognition in speech encompasses a gamut of techniques starting from mechanical recording of audio signal to complex modeling of extracted patterns. Most challenging part of this research purview is to classify the emotion of the speech purely based on the physical characteristics of the audio signal independent of language of speech. This paper focuses on the predictive modeling of audio speech data based on most viable feature set extraction and deployment of these features to predict the emotion of unknown speech data. We have used two most widely used classifiers, a variant of CART and Naïve Bayes, to model the dynamics of interplay of crucial features like Root Mean Square (RMS), Zero Cross Rate (ZCR), Pitch and Brightness of audio signal to determine the emotion of speech. In order to carry out comparative analysis of the proposed classifiers, a set of experiments on real speech data is conducted. Results clearly indicate that decision tree based classifier works well on accuracy whereas Naïve Bayes works fairly well on generality.

Objective Stuttering is a developmental disorder of speech fluency with unknown causes. One of the proposed
theories in this field is deficits in speech motor control that is associated with damaged control, timing,
and coordination of the speech muscles. Fundamental frequency, fundamental frequency range, intensity,
intensity range, and voice onset time are the most important acoustic components that are often used for
indirect evaluation of physiological functions underlying the mechanisms of speech motor control. The purpose
of this investigation was to compare some of the acoustic characteristics of speech motor control in
children who stutter and children who do not stutter.
Materials & Methods This research is a descriptive-analytic and cross-sectional comparative study. A total of
25 Azari-Persian bilingual boys who stutter (stutters group) and 23 Azari-Persian bilinguals and 21 Persian
monolingual boys who do not stutter (non-stutters group) in the age range of 6 to 10 years participated in
this study. Children participated in /a/ and /i/ vowels prolongation and carrier phrase repetition tasks for the
analysis of some of their acoustic characteristics including fundamental frequency, fundamental frequency
range, intensity, intensity range, and voice onset time. The PRAAT software was used for acoustic analysis.
SPSS software (version 17), one-way ANOVA, and Kruskal-Wallis test were used for analyzing the data.
Results The results indicated that there were no significant differences between the stutters and non-stutters
groups (P>0.05) with respect to the acoustic features of speech motor control .
Conclusion No significant group differences were observed in all of the dependent variables reported in this
study. Thus, the results of this research do not support the notion of aberrant speech motor control in children
who stutter.

A common approach in mapping a signal to discrete events is to define a set of symbols that correspond to useful acoustic features of the signal over a short constant time interval. This paper proposes a Hidden Markov Models (HMM) based speech recognition by using cepstrum feature of the signal over adaptive time interval. First pitch period is detected by dyadic wavelet transform and divides the voiced speech signal according to the detected period. Then, system performs HMM-based speech recognition using cepstrum feature to classify the speech signals. Two speech recognition systems have been developed, one is based on constant time framing and the other is adaptive framing. The results are compared and found that adaptive framing method shows better result in both data distribution and recognition rate. ©2009 IEEE.

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5395792