Search

A frequently used procedure to examine the relationship between categorical and dimensional descriptions of emotions is to ask subjects to place verbal expressions representing emotions in a continuous multidimensional emotional space. This work chooses a different approach. It aims at creating a system predicting the values of Activation and Valence (AV) directly from the sound of emotional speech utterances without the use of its semantic content or any other additional information. The system uses X-vectors to represent sound characteristics of the utterance and Support Vector Regressor for the estimation the AV values. The system is trained on a pool of three publicly available databases with dimensional annotation of emotions. The quality of regression is evaluated on the test sets of the same databases. Mapping of categorical emotions to the dimensional space is tested on another pool of eight categorically annotated databases. The aim of the work was to test whether in each unseen database the predicted values of Valence and Activation will place emotion-tagged utterances in the AV space in accordance with expectations based on Russell’s circumplex model of affective space. Due to the great variability of speech data, clusters of emotions create overlapping clouds. Their average location can be represented by centroids. A hypothesis on the position of these centroids is formulated and evaluated. The system’s ability to separate the emotions is evaluated by measuring the distance of the centroids. It can be concluded that the system works as expected and the positions of the clusters follow the hypothesized rules. Although the variance in individual measurements is still very high and the overlap of emotion clusters is large, it can be stated that the AV coordinates predicted by the system lead to an observable separation of the emotions in accordance with the hypothesis. Knowledge from training databases can therefore be used to predict AV coordinates of unseen data of various origins. This could be used to detect high levels of stress or depression. With the appearance of more dimensionally annotated training data, the systems predicting emotional dimensions from speech sound will become more robust and usable in practical applications in call-centers, avatars, robots, information-providing systems, security applications, and the like.

Regional and foreign accents are known to cause problems in many areas of the automatic speech processing, such as speech recognition, speaker recognition, emotion or stress detection. To enable using regional-accent-specific acoustic and language models, the accent of the speaker has to be identified. In this work a preliminary analysis is done on the distinguishability of the Slovak dialects by a basic automatic classifier. In order to determine the range of differences between the Slovak dialects, to estimate the possible extreme values of recognition rates with a given recognizer, the experiments were done on the recordings of speakers using the authentic regional dialects, not just a slightly local-accented standard pronunciation. The reliability of dialect identification, of course, depends greatly on the definition of dialectal areas and on the choice and representativeness of the recordings used for training. We introduce The Sound Archive of the Slovak Dialects. From this collection, we have chosen an appropriate set of recordings and created a database, for dialect-specific acoustic models training and testing. A model of the standard Slovak needed for comparison was trained on the recordings of the Slovak Parliament. A standard Gaussian Mixture Model based recognizer was used for the classification experiments. The results have proven the ability of the GMM classifier to identify the three macro-dialect groups and standard Slovak with reasonable reliability which is already high enough for practical applications. However even better results could be achieved with a bigger volume of suitable representative training data, with gender-dependent modeling, additional features representing prosody, as well as introducing deep-learning modeling techniques, which we consider as future steps.

This document describes the concept of an air traffic management security system and current validation activities. This system uses speech analysis techniques to verify the speaker authorization and to measure the stress level within the air-ground voice communication between pilots and air traffic controllers on one hand, and on the other hand it monitors the current air traffic situation. The purpose of this system is to close an existing security gap by using this multi-modal approach. First validation results are discussed at the end of this article.

In our series of experiments we study weaknesses of the voice biometric systems and try to find solutions to improve their robustness. The acoustical features that represent human voices in the current automatic speaker verification systems change significantly when the person’s emotional arousal deviates from the neutral state. Speech templates of a given speaker used for enrollment are generally recorded in a neutral emotional state using "normal" speech effort. Therefore speaking with higher or lower voice tension causes a mismatch between training and testing resulting in a higher number of verification errors. The acoustical cues of increased emotional arousal in speech are highly non-specific. They are similar to those of Lombard speech, warning and insisting voice, emergency voice, extreme acute stress, shouting, and emotions like anger, fear, hate, and many others. As the available spontaneous emotional speech databases do not cover the full range of the emotional arousal for individual voices, and do not have enough utterances per speaker, we decided to use our CRISIS acted database containing speech utterances at six levels of tense emotional arousal per speaker. Sensitivity of the state of the art i-vector based speaker recognizer with PLDA scoring to arousal mismatch was validated. The speaker verification system was successfully implemented in the online “Speaker authorization” module developed in the frame of the European project Global ATM Security Management (GAMMA). It has been observed that at extreme arousal levels the reliability of the verification decreases. Mixed enrollments with various levels of arousal were used to create more robust models and have shown a promising improvement in the verification reliability compared to the baseline.

The mel cepstral coefficients representing the magnitude spectrum and the Teager energy operators, are often used as features in emotion and stress recognition. The phase spectrum information is generally ignored. In this work we propose the usage of an all pole group delay (APGD) function to represent the phase information, for tense arousal level recognition from speech, along with mel-frequency cepstral coefficients (MFCC) and Critical Band Based Teager energy operator Autocorrelation Envelope features (TEO-CB-Auto-Env). The combination of APGD, MFCC and TEO-CB-Auto-Env features has shown the best recognition results confirming the hypothesis that the phase and magnitude spectra contain complementary information and also their combination with nonlinear features can improve the reliability of the tense arousal level recognition system. The evaluation was performed on the level of whole recordings and we observed statistically very significant absolute improvement of recognition rate of 4.54 % and 4.6 %...

The most commonly adopted approaches in speech emotion recognition (SER) utilize magnitude spectrum and nonlinear Teager energy operator (TEO) based features while information about phase spectrum is often omitted. The information about phase has been frequently overlooked in approaches applied by speech processing researchers due to the signal processing difficulties. We present study of two phase-based features: The relative phase shift (RPS) based features and modified group delay features (MODGDF) that represents phase structure of speech in the task of emotional arousal recognition. The evaluation is performed on the CRISIS acted speech database which allows us to recognize five levels of emotional arousal from speech. To exploit these features, we employ concept of deep neural network. The efficiency of the approaches based on features mentioned earlier is compared to baseline platform using Mel frequency cepstral coefficients (MFCCs) and all pole group delay features (APGD). The combination of anot...

The voice communication between pilots and the air traffic controllers is vulnerable to various types of attacks. Speaker verification could be used as an add-on security feature; however, there are several factors that make the use of voice biometry in this scenario difficult to apply. These are among others: open set of speakers, very short utterances, speaker noises, signal clipping, foreign accent of non-native speakers, and high content of background and channel noises in the signal. This paper identifies sources of noise in the entire communication channel and analyzes the influence of these noise components of different types and levels on the reliability of the speaker verification. An i-vector based speaker recognizer with PLDA scoring is used for the experiments. Cockpit noises of several aircrafts and limited-band channel noises are simulated by a software noise-generator. The sensitivity of the speaker verification to the noises of different frequency bands is studied in comparison to the long...

The mel cepstral coefficients representing magnitude spectrum and Teager energy operators are often used as features in emotion recognition. The phase spectrum information is generally ignored. In this work an approach is proposed based on the use of group delay function from all pole models (APGD) to represent the phase information for the emotional arousal recognition from speech. The experiments were done on the CRISIS acted speech database with four levels of stress. The results of the arousal recognition system using the APGD features are compared to those using mel-frequency cepstral coefficients (MFCCs) and with Critical Band Based TEO Autocorrelation Envelope features (TEO-CB-Auto-Env) which have been successfully used in the task of emotion and stress detection in the past. The feature extraction is applied on the voiced parts of speech. The combination of APGD, MFCC, and TEO-CB-Auto-Env features has shown the best recognition results confirming the hypothesis that the phase and magnitude spectra...

Puppeteering represents a traditional but slowly dying folk art in which impersonation based on modifying the puppeteer's voice belongs among the core expressive and stylistic devices. A study of the formal and functional aspects of voice quality and other prosodic features in puppeteers' production as well as their acoustic correlates is thus an integral part of studies into communicative social signals especially in emotional and expressive speech. This analysis becomes even more relevant in cases when puppeteers' performance represents a very salient communication code in a particular culture. Better understanding of such traditional code opens possibilities of applications into several speech processing domains. Our data consist of recordings of plays and excerpts by two puppeteers (a father and a son) and verbal descriptions of personality features of the characters that the actor tries to express through their voices. We propose an approach to the taxonomy of archetypal characters based on analyzing the psychological, aesthetic and acoustic-phonetic aspects of their personalities. More specifically, we identify salient links between acoustic-prosodic features on the one hand and semantic and personality features on the other hand.

This study is an extension of previous research on so-called "low-level" linguistic units in general, and on Slovak letter and grapheme frequencies in particular. Specifically, the ranking behavior of Slovak letters and graphemes is compared to the organization of allophones, submitting identical linguistic data to analyses on all three levels. As a result, it can be shown that all three kinds of units share some common features as to their frequency organization; still, there seem to be substantial differences. As to common traits, it is shown that the frequency of Slovak letters, graphemes and allophones can be modeled by the negative hypergeometric distribution; furthermore, there is a linear relation between parameters K and M of this model, previously observed and interpreted in more detail for other Slavic languages. However, as opposed to previous results obtained, parameter K does not seem to depend on inventory size only; also, whereas there is no significant difference of (relative) entropy between LG frequencies, the difference is significant between both and allophones, (relative) entropy being less for A frequencies as compared to the LG frequencies. These findings can be interpreted in terms of a more uniform exploitation of allophones as compared to that of letters or graphemes, indicating a relative under-exploitation and/or over- exploitation of LG units as compared to the allophone inventory. Perspectives for further research are outlined.

Nasality, nasal sound, nasalized sound — these expressions are used in phonetics and speech research, in singing theory, and in musical acoustics. In this paper we try to find a classification for several different types of nasality and to show links between the meaning of the expression nasality in speech, singing, and musical acoustics. A number of spectral measurements, listening experiments and discussions were undertaken to find relationships for nasality in these three fields.

Bagpipes are the only group of traditional folk reed instruments that have survived in Slovak folk culture until now. Unlike Scottish highland pipe, gaita, cornemuse etc., all the types of Slovak bagpipes use single reeds in both chanters and drones. Moreover the horns of the drones are partly closed with a plate having only a small opening for air output and sound radiation. This forms a Helmholz‐like resonator ‐ acoustic filter having a strong influence on the acoustic and aesthetic properties of the sound. The ideal of the sound timbre of these pipes is "less open" (compare to open and closed vowels in speech) and, in spite of its relatively high loudness, "less agressive" in comparison to pipes with open horns. This paper presents our approach to study the ideal of sound timbre of the Slovak bagpipe using a source‐filter all‐pole model. To identify resonance frequencies/formants, we applied inverse filtering technique using discrete linear prediction. Some properties of the source (reed) were studied using parameters like amplitude qutient and open quotient. The possible relations between timbre of speech sounds (vowels) and the timbre of the sound of musical instruments are discussed.

The paper describes methodology for creating a Slovak database of speech under stress and pilot observations. While the relationship between stress and speech characteristics can be utilized in a wide domain of speech technology applications, its research suffers from the lack of suitable databases, particularly in conversational speech. We propose a novel procedure to record acted speech in the home of actors and using their own smartphones. We describe both the collection of speech material under three levels of stress and the subsequent annotation of stress levels in this material. First observations suggest a reasonable inter-annotator agreement, as well as interesting avenues for the relationship between the intended stress levels and those perceived in speech.

Presence of appropriate acoustic cues of affective features in the synthesized speech can be a prerequisite for the proper evaluation of the semantic content by the message recipient. In the recent work the authors have focused on the research of expressive speech synthesis capable of generating naturally sounding synthetic speech at various levels of arousal. Automatic information and warning systems can be used to inform, warn, instruct and navigate people in dangerous, critical situations, and increase the effectiveness of crisis management and rescue operations. One of the activities in the frame of the EU SF project CRISIS was called "Extremely expressive (hyper-expressive) speech synthesis for urgent warning messages generation''. It was aimed at research and development of speech synthesizers with high naturalness and intelligibility capable of generating messages with various expressive loads. The synthesizers will be applicable to generate public alert and warning messages in case of fires, floods, state security threats, etc. Early warning in relation to the situations mentioned above can be made thanks to fire and flood spread forecasting; modeling thereof is covered by other activities of the CRISIS project. The most important part needed for the synthesizer building is the expressive speech database. An original method is proposed to create such a database. The current version of the expressive speech database is introduced and first experiments with expressive synthesizers developed with this database are presented and discussed.