The application of laryngograph in research of quality of speech signal. The electroglotography method

Transcription

1 The application of laryngograph in research of quality of speech signal. The electroglotography method JOLANTA ZIELIŃSKA Department of Special Pedagogy Pedagogical University of Cracow Ingardena Kraków, Poland jzielin@op.pl EWA BRZDĘK Department of Special Pedagogy Pedagogical University of Cracow Ingardena Kraków, Poland ebrzdek@wp.pl Abstract: - The elecroglotography method is an instrumental analysis by the device called Laryngograph, which is presented as a practical application in the speech signal analysis issues. In this paper capabilities of this device are estimated. A very interesting fact is that visualization of the speech signal obtained with the use of Laryngograph device allows to detect it s acoustically and phonetically most important features and present them in a graphical form. The analysis process performed with use of computer and presented computer attachment is easier, faster and ensures higher quality than other methods. Computer voice recording enables not only visualization but also objective assessment and its repetitiveness. In the context of presented questions practical capabilities of integrated system for speech examination (Speech Studio) are discussed. Key-Words: - computer visualization, speech signal, electroglotography, laryngograph, voice profile, vocal cords irregularity. 1 Introduction The speech analysis is a very complicated process because it is concerning highly complex signal. This is the main reason that it is in the area of interest of specialists from many scientific disciplines, so varied as linguistics [2, 8, 12, 14] and computer science [3, 5, 10, 13]. Acoustic phonetic concerns the technical-physical aspect of the speech signal. It analyzes the speech signal, as a sound realization of language code, with use of properly extended acoustic system [4, 7, 9]. The very important aspect of high-tech phonetic research is an instrumental analysis of speech signal with use of such devices as: x-ray machine, electromyograph (EMG), spectrograph, minograph, Laryngograph. Its goal is speech signal visualization, which makes possible to detect its most important features and present them in a graphical form. The analysis process performed with use of computer is easier, faster and ensures higher quality than other methods. Computer voice recording enables not only visualization but also objective assessment and its repetitiveness. These are the main advantages of the electroglotopraphy method. 2 Electroglotography method general characteristic The electroglotography method is based on the electrical impedance measurement between vocal folds during their vibrations during speaking process. It s a function of their mutual position and changes during larynx vibration. When vocal folds are cramped the value of this function is smaller, than in the case when they are opened. The monitoring of vocal folds movements during normal speaking process takes advantages of this phenomenon. However there are some disadvantages of devices, which use it directly. Individual regulation for each patient is necessary and some distortions can appear, because the electrodes, which are used for impedance measurement, can translocate on the patient throat. The technique, which presents the improved form of electroglotography and eliminates mentioned disadvantages is called loryngographical and was practically introduced in Laryngograph device. The signal obtained with use of elegctroglotography, which represents the functioning of vocal folds, establishes the basis for Fx basic frequency determination (presented in figure 1). Fx frequency is very important in the phonetic research. ISBN:

2 Phonetics of University College in London [1]. Nowadays it is produced in Laryngograph Ltd. Company. During several years it was put into practice in such domains as: Fig.1 Sample curve of Lx signal determination of basic Fx frequency [15] The initial signal of the Lx (larynx) signal the functioning of vocal folds and is the base for determination of basic frequency Fx. Analysis of the signal cycle Lx helps to separate its three phases. The first phase is a quick signal growth, corresponding to the quick closing of the vocal folds. The second phase corresponds to the slightly slower signal fall and is connected with their opening. The third phase, which is a flat waveform, corresponds to the state, when the vocal folds are open. Fx frequency is calculated while measuring of each functioning cycle length in the middle of the Lx signal amplitude value (in practice, the measurement of time intervals between transitions of the signal through that level is made). The sought Fx value is the reverse of Lx signal cycle length. 3 Practical usage of electroglotography method usage of coputer attachment Laryngograph processor PCLX Computer attachment Laryngograph Processor PCLX operation is based on the electroglotography method. It is connected to PC computer with use of interface card, installed directly on IBM PC computer bus, which constitutes coupler with Laryngograph and software. This card contains special system, which processes signals and makes it possible to visualize them on the computer screen and remember just in-time. PCLX software called Speech Studio allows signals activation, visualization, on-line analyze and remembering them on the hard drive for later analysis and recovery. The prototype of presented device was introduced for the first time at the Faculty of speech therapy, in order to improve speech process; clinical examination of nose and throat, to diagnose work of vocal folds after surgical procedures; improvement of singers and announcers voice, examination relations between speech and work of vocal cords; teaching foreign languages. The experimental research was made in Cracow (Poland) by inż. dr hab. J. Zielińska [15]. This investigation concerned the evaluation and the development of oral ability in deaf children. The aim of the research was the assessment, evaluation and positive changes in the level of oral development in deaf children reception hearing system. The research have been conducted with the use of two computer attachments: Laryngograph Processor PCLX and Nosality, connected to a PC computer. The obtained results and following practical conclusions can serve as guidelines in clinical and logopedical applications in voice rehabilitation and communication development of groups of non-hearing children whose native language is not necessarily Polish. This kind of computer research is the first and unique. It fully considers the specific character of the Polish language, including very difficult dental phones, digraphs and prosodic features of speech, mainly the flow of the melodic waveform, which largely influences the understanding of speech in Slavonic languages. Hence, the research results can be applied to not only diagnosis and rehabilitation of children with oral problems, for whom the verbal language is often not a natural language, i.e. those children who cannot hear, but also for those, who have cleft palates or other defects. They can also help teach Polish to foreign students. The most important element of the speech signal analysis is the way the signal is sampled and remembered by computer. Computer attachment Laryngograph Processor PCLX possessed three input signal detectors. These are coopery electrodes placed on the patient s throat with use of regulative band. Necessary condition of acquiring correct research result is the way that electrode are placed on the patient s throat. It depends on the age of person: if the person in younger, electrodes should be placed higher, assuming central location between ISBN:

3 4-rd and 7-th neck vertebra in case of adults. This fact comes out directly of the physiological structure of the larynx [3, 4, 7, 15]. At that age it is important to pay special attention to physiologically correct placement of laryngograph electrodes. Fig. 2 Computer presentation of signal mama (mother), enabling the evaluation of child s oral efficiency [15] The initial signal of the laryngograph Lx (the second waveform in figure 2) reflects the work of vocal folds and is the base for determination of basic frequency Fx (waveform in figure 2). Analysis of the signal cycle Lx helps to separate its three phases (closing, connecting, opening of vocal folds). Fx frequency is calculated while measuring of each work cycle length in the middle of the Lx signal amplitude value (in practice, the measurement of time intervals between transitions of the signal through that level is made). The sought Fx value is the reverse of Lx signal cycle length. The shape of Lx signal gives information about any abnormality of vocal route work, especially the work of vocal folds. The second decoder (microphone) should be placed about 2,5cm from the patient s mouth (line Sp in figure 2). Sp signal (the first waveform in figure 2) which is obtained form the microphone inlet enables a dynamic visualization of signal timing, evaluation of its volume, loudness and exercising dentalised phones, i.g. pairs s sz, z- rz, which are very often misused and mistaken by persons having oral problems, including lisping. The analysis of Fx signal gives information about all parts of a proper signal, which are easily evaluated even by a child. They are: proper dynamic breathing, economical breathing, work of vocal chords, proper articulation, prosodic features, such as: speed, rhythm, accent or melody, i.e. these elements, which decide about signal expression and, additionally show the level of phone voicing (no voicing no picture). The third electrode placed on nostrils is its inlet. The waveform generated by this electrode on computer screen shows the dynamics of the air flow through the nose of the examined person, by measuring nostrils vibrations, like an accelerator which measures changes of speed in time, i.e. the acceleration of nostrils movement (lines NxACC in figure 2). Precise maximum amplitude values can be specified by creating a two-dimensional amplitudefrequency spectrum. PCLX software of the Laryngograph chart has these possibilities. Maximum values, visible on computer screen are called formants. Their frequency and amplitude characterize a phone and are closely related to articulator configureuration that has appeared during its creation. Figure 2 is a deliberately transformed picture for the fullanalysis of oral efficiency. NxACC signal (third waveform in figure 2) enables to determine proper nasal phones production, including the work of glottis. Such analysis of voice parameters enables to draw many practical diagnostic-therapeutic conclusions. 4 Analysis Of Speech Signal With Use Of Computer Attachment Laryngograph Lx At present we are investigating the following issues: - the general profile of speech patterns, - the vocal cords work irregularity. The research work is conducting in the team of patients with dysphonic problems and voice disorders, for examples: laryngitis, chorditis, vocal cord nodules, vocal fold cyst, Reinke s Edema, laryngeal papillomatosis, gastroesophageal reflux disease, laryngeal cancer, myasthenia gravis. During voice generation process three basic phases take place and they can be fully monitored and analyzed by computer software called Speech Studio. These three phases are: generation of stimulated vibrations, its modification in vocal resonator and radiation [3, 13]. The signal which is generated in larynx during voiced sound pronunciation, by vocal cords in the role of stimulation generator, is a periodic flow, characterized by the basic frequency Fx, which was described in details earlier (figure 1). In the narrowness of the voice channel the signal is modulated. The adjustment of voice route, so ISBN:

4 articulators configuration, is characterized by formants, which are defined on based of average amplitude-frequency spectrum (presented in figure 4). Figure 3 presents the main screen of integrated system for voice examination Speech Studio, which have all the possibilities of speech signal analysis that were described previously. During work with this program there are two toolbars available (top and side one), which allows to perform several operations, starting with signal activation, its remembering, recovering, analysis, till configuring the amount and type of visible flows and scales. values, which characterize the speech profile of the patient. In this profile there are defined following parameters: minimum, maximum, average value, standard deviation for the basic Fx frequency and occlusion factor Qx (WZG) and also other parameters describing disturbances of the frequency (Jitter) and amplitude (Shimmer). The flow of the occlusion factor Qx depends directly on the Lx flow and is presented as a percentage value. It determinates the percentage usage of time, when vocal folds are opened, that is the time of occlusion and opening during statement pronunciation (the sum of phase 1 and phase 2 durations in figure 1). During this type of research, related to the speech signal analysis on the computer screen there are simultaneously presented many flows, which characterizes the speech signal. On base of these flows we can define both the numerical parameters describing the voice during appropriate analyses and qualitative conclusions. This program allows reducing the number of presented charts to the selected once. Presented speech signal spectrum is obtained with appropriative computer procedures, which are based on FFT algorithm (Fast Fourier Transformation) and LPC (linear prediction algorithm). It allows to determinate formants in selected period of time and the momentary spectrum in specific point of time. For each format its maximum frequency value, amplitude and frequency band is presented. Fig. 3 Integrated system for voice examination Speech Studio [15] The fundamental part of the screen is seven flows, which characterizes the speech signal. The two most important flows present changes of amplitude in time and the work of vocal folds. These charts can be enlarged in the bottom right window and allows to immediate preliminary diagnose the voice pathology and correct adjust equipment parameters. Signals amplification can be regulated by attachment knobs or sliders placed in the bottom left corner of the screen. In the bottom part of the screen there are also two other windows. In the first one average amplitude-frequency spectrum are displayed. In the second one there are 5 Computer Analysis Of Research Material There are various possibilities of research material analysis starting with statistical calculations, parametrical calculations, till graphical processing. In order to obtain more precise signal view we can select any part of it and freely enlarge, so that even the smallest pathological change can be detected. This software enables also measurements of time intervals between individual points of the signal. It can be performed by appropriative placing two cursors on the screen and time interval between them will be measure. The Fx signal analyses is based on determinations of 23 histograms, including CFx histogram, which presents the vocal cords work irregularity (figure 4). ISBN:

5 Fx 2 (H z) Fx1 (Hz) Samples = 128 Irregularity = 2.34% < > = 0, 0 Fx1 (Hz) = * Fx2 (Hz) = * CFx C:\SpeData\LX\mama.spe Fig. 4 Sample CFx histogram, which presents the vocal cords work irregularity [15] Computer analysis of speech signal allows obtaining the voice profile of the patient in the form of acoustic parameters (figure 5). Fig. 5 Sample voice profile for mama statement, Speech Studio program [15] The voice profile contains the average value of basic Fx frequency, its minimum and maximum value. It makes also possible to determinate frequency oscillations, standard deviation, the factor of frequency disturbance called Jitter, percentage value of the average occlusion factor Qx, its minimum and maximum value and standard deviation. Additionally, as the last information, the value of L x voice amplitude disturbance factor (called Shimmer) is presented. 6 Conclusion The usage of electroglotography method enables to deeply analyze the speech signal, which was remembered in the computer memory. For example it can be a general spectral analysis or detailed analysis based in of their formants. By means of basic Fx frequency visualization on the computer screen it is possible to estimate the voice pitch as well as to obtain the full acoustic profile of voice, with use of instrumental analysis. The general conclusion is that as complex as the speech signal can be also methods of its examination. The technique must follow this track. The electroglotography method, presented in this paper, is a proof. The scope of its usage depends mainly on the designated research goal and competence of the person performing the examination. The undisputed fact is that presented opportunities are really extensive and, what is the most important fact, comparatively easy in obtaining and interpretation. References: [1] E. Abberton, D. Howard, A. Fourcin, Laryngograph assessment of normal voice: a tutorial, Clinical Linguistic and Phonetics 3, Informa Healthcare 1996, pp [2] E. Abberton, Phonetic consideration in the design of voice assessment material, Logopedics Phoniatrics Vocology 30, 2005 pp [3] E. Abberton, Aspects of voice quality in women, Measurement of Speech Sound Data and its Practical Application, 2002, pp [4] E. Abberton, Voice quality of deaf speakers, (in:) Kent, R. D., Ball, M. J. (eds.) Voice quality measurement, Singular Publishing 2000, pp [5] E. Abberton, E. Carlson, A window on the voice - how I use computers in therapy, Speech and Language Therapy in Practice, 1999, pp [6] C. Basztura, Źródła, sygnały i obrazy akustyczne (in Polish), Wyd. Komun. i Łączn., Warszawa, [7] P. Cheung, E. Abberton, Patterns of phonological disability in Cantonesespeaking children in Hong Kong. International Journal of Language and Communication Disorders 35(4), 2000, pp ISBN:

6 [8] G. Demenko, W. Jassem, Phonetic and syntactic coherence of the phrase, (in:) W. Jassem, C. Basztura (eds), Speech and Language Technology, vol. I, Wrocław: Format, 1997, pp [9] A. Fourcin, E. Abberton, Hearing and phonetic criteria in voice measurement: Clinical applications, Logopedics Phoniatrics Vocology 33(1/2008), pp [10] I. Guimaraes, E.Abberton, Fundamental frequency in speakers of Portuguese for different voice samples, Journal of Voice 19(4), 2005, pp [11] I. Guimaraes, E. Abberton, Health and voice quality in smokers: an exploratory investigation, Logopedics Phoniatrics Vocology 30(3), 2005, pp [12] M. Karpiński, Struktura i intonacja polskiego dialogu zdaniowego (in Polish), Wydawnictwo Naukowe UAM, Poznań [13] R. Tadeusiewicz, Sygnał mowy (in Polish), WKiL, Warszawa, [14] M. Taniguchi, E. Abberton, Effect of interective visual feedback on the improvement of English intonation of Japanese EFL learners. Speech, Hearing and Language: work in progress 11, 1999, pp [15] J. Zielińska, Diagnoza i terapia sprawności ortofonicznej dzieci z uszkodzeniem słuchu wspomagane techniką komputerową (in Polish), Wyd. Naukowe AP, Kraków, ISBN: