|
|
Implementation of 1D mathematical model of vocal cavities into TTS synthesizer – preliminary study
Radolf, Vojtěch ; Horák, Petr
Simplified 1D mathematical models of the human vocal tract were modified for using them in Text-To-Speech systems so that they help to simulate emotional speech. The geometry (area function) of the models for all Czech vowels was modified using the inverse task optimization procedure so that the computed formant frequencies match the measured formant frequencies of utterances of professional speaker. Output acoustic pressure signal generated from the models in wav format sounded satisfactorily for all the vowels and fundamental frequencies varied in an octave range from 77 Hz to 156 Hz. Neverthelles more testing procedures are needed to verify reliability and quickness of the model as well as intelligibility of generated utterances especially in formant TTS system and linear predictive TTS system.
|
|
|
Measurement of contact stress in a self-oscillating model of human vocal folds
Horáček, Jaromír ; Bula, Vítězslav ; Veselý, Jan ; Radolf, Vojtěch
The contribution presents in vitro measurement of contact stress in the artificial vocal folds made of a silicon rubber excited by airflow with synchronous registration of the flow induced vocal fold vibrations using a high speed camera, measurement of subglottal dynamic and mean air pressure and the generated acoustic signal. The measured maximum impact stress, maximum glottal opening and sound pressure level are compared with data found in excised larynges as well as with the values numerically simulated by the aeroelastic model of vocal fold self-oscillations.
|
|
|
Experimental investigation of air pressure and acoustic characteristics of human voice. Part 1: Measurement in vivo
Horáček, Jaromír ; Radolf, Vojtěch ; Bula, Vítězslav ; Veselý, Jan ; Laukkanen, A. M.
This contribution is aimed to provide material that can be used to develop more realistic physical models of voice production. The experimental methodology and the results of measurement of subglottal, oral (substitute for subglottic) and acoustic air pressure (captured at a distance of 20 cm in front of the subject’s mouth) are presented. The data were measured during ordinary speech production and when the acoustic impedance and mean supraglottal resistance were raised by phonating into differently sized tubes in the air and having the other end submerged under water. The results presented in time and frequency domain show the physiological ranges and limits of the measured pressures in humans for normal and extreme phonation.
|
|
|
Computer simulation of musical singer’s voice based on MRI and acoustic measurement
Radolf, Vojtěch ; Nissinen, A. ; Laukkanen, A. M. ; Havlík, R. ; Horáček, Jaromír
An inverse method was used to estimate the vocal tract geometry as a 1D model on the basis of acoustical characteristics of a professional musical singer before and after vocal exercising. The basic geometrical data for the model were obtained from MRI registered during sustained phonation of vowels [a:], [i:], [u:] produced in naive and professional ways. The model was used for numerical simulations of the voice signals. The results of simulation were compared to the acoustic recordings. According to the results, a singer’s formant cluster was accomplished after exercising. It seemed to be due to lowering of the larynx and lengthening and narrowing of the epilarynx. There was a qualitative agreement between 3D measurements of MRI and the results of modeling. The results suggest that for a singer’s formant cluster a relatively low pharynx over epilarynx ratio may be sufficient, at least if the larynx lowers.
|
|
|
Analysis of the influence of piriform sinus on vocal tract acoustic characteristics
Radolf, Vojtěch
An analysis of several testing mathematical models and human vocal tract model with piriform sinus (PS) cavities was done. The method used was based on 1D acoustic wave equation, considering viscous losses, radiation impedance losses and applied on conical elements in frequency domain. The analysis revealed that the 1D model is able to capture differences in acoustic variables of the vocal tract modified with PS. Nevertheless the model includes PS of very small dimensions (9 mm in depth) that generate spectral changes far away from frequency band 4-5 kHz as Dang&Honda (1996) observed with their models including PS up to 20 mm in depth.
|
|
| |
|
|
1D modeling of the influence of velopharyngeal insufficiency on phonation of vowels
Radolf, Vojtěch ; Vampola, T.
Velopharyngeal insufficiency is modeled in frequency domain by the transfer matrix method in connection with conical acoustical elements. The vocal tract is considered as a branched system with two nasal ducts. The influence of viscous losses and radiation impedance both at the level of the lips and at the level of the nostrils are considered. Configuration of the vocal tract corresponds to the vowel /a:/. The modal analysis is performed and the results are compared with those gained from 3D FEM models.
|
|
| |
|
|
PROBLEMS OF SINGER’S FORMANT MATHEMATICAL MODELING
Radolf, Vojtěch
Theme of the paper belongs into the field of acoustics in biomechanics. In co-operation with foniatric departments in Czech Republic the spatial and then simplified geometric models of acoustic human vocal cavities were created. Using various mathematical models the dynamic characteristics of the models were finding out. One of the objectives of the project is to investigate singer's formant cluster. The mathematical models are designed to find such configuration of acoustic cavities, respecting real physiological limits, which leads to excitation of predefined acoustic resonances. It is so called formant tuning. Using these models the appropriate geometry for the singer's formant cluster formation were searching for.
|
|
|
Optimization of resonance properties of the human vocal tract by variation of cross-section and length of acoustic elements
Radolf, Vojtěch
Theme of the paper is focused on optimization possibilities of geometrical form of human vocal tract. An optimization process is designed to find such configuration of acoustic cavities, respecting real physiological limits, which leads to excitation of predefined acoustic resonance. Especially frequency domain between the third and the fifth formant is important for voice timbre, namely for singing voice ('the singer's formant'). Two various 1D optimization methods, computing time and accuracy of solution are compared. The first method is a general procedure of numerical searching for the minimum of objection function, the second is an energetic iterative procedure using non-dimensional sensitivity functions. Both length and cross-section area of conic elements are varied.
|
guest ::
