MDVP definition

Sustained Sound MeasurementsPerturbation = average difference between successive vocal fold cycles (1)Equation (1) provides a straightforward basis in words for thecharacterisation of perturbation in a way that directly relates to the concept and that can be used for the whole of a selected portion of a vowel sample. Equation (2) is a more formal definition that lends itself to calculation for the N cycles of the selected sample. (Note that the selection is ordinarily made so that the onset and offset of the vowel are avoided [Titze, (1995), and Baken & Orlikoff (2000) give very useful discussions.)Perturbation magnitude =∣-∣-+-=∑n n N n x xN 11111(2)When x n = peak pressure amplitude, Pa, in the n th vocal fold cycle of vibration Shimmer magnitude in Pascals is defined [*].When x n = duration, ms, of the n th vocal fold cycle of vibrationJitter magnitude in milliseconds, is defined [*].When logarithmic units are used these expressions for average shimmer and average jitter represent relative values [*]. This is because they are then based on ratios of pressure and of cycle duration respectively, and, as ratios, they become dimensionless.Shimmer dB =∣-∣-+-=∑dB )(dB )(11111n n N n x x N (3)Jitter semitones, st =∣-∣-+-=∑st )(st )(11111n n N n x x N (4)The use of dB and semitones is appealing because the resulting measures are more familiar and are intrinsically relative rather than absolute; they are better correlated with the corresponding sensations. In practice, however, they are little used and the need to be able to use relative values has led to the much more widespread employment of simple ratios. The description in words then becomes:Relative perturbation =cycle fold vocal average cyclesfold vocal successive between difference average (5)and more formally for the purposes of calculation:Relative perturbation = ∑∑==+-=∣-∣-N n n n n n N n xN x x N 1111111 x 100% (6) Shimmer is defined whenx n = peak pressure amplitude in the n th vocal fold cycle of vibrationShimmer+ (Laryngograph) is obtained by using the peak pressure following vocal fold closure in each cycle and then multiplying (6) by 100 to get the percentage value. (This technique provides the most accurate basis for the calculation — compared with acoustic approaches.)Shimmer- (Laryngograph) is derived in the same fashion but using the negative peak pressure value following closure.Jitter is defined whenx n = fundamental frequency of the n th vocal fold cycle of vibration%Jitter , the most often used version of this parameter, is then obtained by multiplying (6) by 100; Jitter First (Laryngograph) is derived in exactly the same way except that, in the interests of greatest accuracy, only the positive going Lx closures are used to signal the beginning of each vocal fold cycle. Jitter Ratio is obtained by multiplying (6) by 1000 (Horii, 1979)Jitter Factor is obtained by using frequency values in (6) and multiplying by 100 (Hollien et al 1973). When the accuracy available from the use of the Laryngograph Lx waveform is obtainable and the degree of jitter is ≤ 1%, there is little difference between the percentage values of Jitter factor andJitter Second (Laryngograph) again uses Lx to define closure instants but equation (6) is modified so that an average over two consecutive cycles is used in the calculation, so as to reduce the effects of any consistent up or down local trend in Fx; x n is the nth cycle fundamental frequency in (7).Jitter second = ()∑∑==-=-++--N n n nN n n nn xN x x x N 1121113221 x 100% (7) RAP — relative average perturbation (Koike, 1973) is obtained by thecalculation of a running average over three consecutive vocal fold cycles to give a measure of jitter that is very similar to Jitter Second and similarly reduces the effect of trends in voice “pitch”, x n is nth cycle duration in (8).RAP = ()∑∑==-=+--++-N n n nN n n n n n xN x x x x N 112111321 x 100% (8)The measurement of “noise” ratiosThe essential notion here is that a pathological voice signal can be regarded as being composed of two components. The first is a well defined periodic voice signal that has a clear harmonic spectrum; the second is an additional source of noise contamination [Yanagihara et al, 1982]. This concept that there may be clinical value in the measurement of noise that may be embedded within pathological voice signals has led to the introduction and use of variousmeasures of “Harmonic to Noise Ratio”. Its efficacy is based, however, on the notion that the speech sample will be stable with negligible variation of period and spectrum through the sample:HNR dB = 10*log 10alone the noise energy in ics alone the harmon tained in energy con (9)This is also a computationally intensive measurement that requires an appreciable time for its completion for an utterance of several seconds duration. The delay with even quite fast PC facilities is clinicallyinconvenient and several other approaches have been adopted. One method is to reduce the number of cycles that are analysed and calculate theNormalised Noised Energy [NNE; Kasuya et al, 1986]; and this has the further advantage that the effects of both vocal fold frequency and of vocal tract movement variation within the sample are minimised.NNENormalised noise energy dB = 10*log 10the signal energy in the noise tained in energy con (10)Signal to Noise RatioAnother approach [Klingholtz, 1987] is simply to avoid the need for harmonic analysis yet effectively achieve the same end r esult by calculating the “signal to noise ratio”:S/N dB = 20*log 10 s raw cycle n mean and ces betwee f differen average o cyclefold vocal mean of amplitude average (11)The underlying thought here is that since noise is random its average is zero and the average amplitude will be an essentially noise free estimate. Subtraction from the raw signal then leaves the noise alone. The process depends, however, on the assumptions that the voice signal is truly periodic, and that vocal tract is held in a steady position.It is not easy perceptually to make the distinction between the sensations associated with: irregularity of duration from cycle to cycle of vocal fold vibration; irregularity of amplitude from cycle to cycle; and the presence of random noise in the voice. Close examination of the acoustic waveform for a pathologically hoarse voice may, for example, reveal quite considerable cycle-to-cycle variability in both amplitude and duration but show little evidence of the random noise component that comes from air turbulence. The percept, however, could easily be taken in error to be one owing its origin to noise rather than to intrinsic waveform variability.Cepstral Peak ProminenceThis technique has its origin in the idea that, since the frequency spectrum of a periodic sound is periodic in frequency, this frequency spectrum can itself be spectrally analysed. The result of this double process of frequency analysis is called a cepstrum. The cepstrum of a periodic signal typically has a prominent peak at a point corresponding to the fundamental period of the sound. A practical advantage of this analysis is that it does not require the temporal definition of individual periods (deKrom, 1993). Hillebrand et al (1994) have applied this approach to the analysis of sustained vowels inpathological speech. By comparing the height of the cepstral peak to the rest of the cepstrum a measure of the intrinsic periodicity of a sound can be defined — the Cepstral Peak Prominence (CPP):CPP dB = 20*log 10 cepstrum of eight average h magnitudepeak cepstral (10)In Laryngograph® VoiceSuite the average height of the cepstrum is calculated by defining and separately averaging the cepstrum level on each side of the first peak.Contact Phase — Qx%Contact phase is used to refer to the estimation of vocal fold contact that is based on the use of the laryngograph , Lx, (egg) waveform. InLaryngograph® VoiceSuite the time width of the closure waveform at 70% of the peak to peak value of the Lx waveform is used. Ordinarily, thepercentage ratio of the “closed phase” to the total period for each vocal fold vibrational cycle is used for quantitative measurement. DQx 1&2 refer to the first and second order distributions for samples of connected speech. CQx is the corresponding period by period cross plot for the whole sample.Overall Timing ControlTotal entropy {voicing; (silence + voiceless sounds)} = Γp(v)logp(v) +Γp(s)logp(s)where p(v) is the probability of voiced sounds and p(s) is the combined probability of voiceless sounds and silence.。