Riede, Tobias : Vocal changes in animals during disorders

51

Chapter 4. The harmonic-to-noise-ratio applied to dog barks

4.1 Introduction

In order to find out which acoustic parameter might be susceptible to changes when the senders state of health is affected, we investigated several case studies and found that the amount of nonlinear phenomena increases under certain pathological conditions in the original harmonic vocalisations, like the Japanese macaques coo-call (Riede et al. 1997), the domestic cats meow-vocalisation (Riede, Stolle-Malorny 1998, 1999) or the wolf-howl (Riede et al. in press).

Vocalization which bear originally a considerable amount of nonharmonic energy (noise) can not be evaluated with this procedure. For instance, the dog bark contains harmonic and non-harmonic energy. Dogs utter the barking sound in different behavioural contexts (Feddersen-Petersen 1996). The dog bark can be considered harmonic to a varying degree, with a considerable amount of aperiodicities at the same time. Vocal changes of the bark, which are for instance described as hoarseness (Bagley et al. 1996) seem to be founded on a shifting of the ratio of the energy of harmonic and non-harmonic elements. Additionally, in normal dogs, some authors assume different communicative relevance according to this ratio (Tembrock, 1976; Feddersen-Petersen 1996, Wilden 1997).

In this study the harmonic-to-noise-ratio (HNR) was adapted, which is a quantitative measure well known in human voice research. The HNR considers the acoustic energy, i.e. the amplitudes, of the harmonic components to that of the noise (Yumoto et al. 1982; Awan, Frenkel 1994). The auditive impression of a voice with a low HNR is overwhelming hoarse, whereas a voice with a very high HNR sounds hypertensive/pressed.

In this study a HNR-calculation procedure was applied to dog barks. The main goal of this study was to introduce the HNR calculation procedure into bioacoustics. For this reason the procedure has to be validated. First we used synthetic sounds with defined HNR values and applied our algorithm, and second the auditive impression of human subjects were compared with the HNR values of the barks. Third the multiparametric analysis, a tool often used in animal bioacoustics, was applied to reproduce the HNR results.

We recorded the barks of dogs expressing some sorts of dysphonia as well as a sample of


52

normal dogs.

4.2 Material and Methods

Studied animals

Audio recordings from ten healthy Dachshund dogs considered as the ’normal sample‘ were taken at the owners property. The dogs ranged in age between 9 months and 11 years and in weight between 6.8 kg and 10 kg. A general clinical investigation of the ten dogs confirmed no peculiarities.

Dachshunds, considered as 'clinic sample', were recorded in the Veterinary Clinic for small animals at the Freie Universität Berlin. During a 6-months period 9 dachshunds admitted in the clinic for thoraco-lumbal intervertebral disc extrusion were considered. One additional dachshund expressed excessive barking during the stay in the clinic and developed a hoarse-sounding voice. Dogs ranged in age between 3 and 8 years and in weight between 7.9 kg and 10.7 kg.

Acoustic recordings

Recordings were made with a Marantz PMD 222 tape recorder and Sennheiser microphone (ME80 head with K3U power module) on BASF ChromeSuper II tapes. Distance between dog and microphone varied between 50 cm and 150 cm. Barking was released by staring into the dog's eyes (a mild threat to the animal). Only animals in which barking could be released in this way were included to the samples.

Barking is usually uttered in sequences of variable duration consisting of barks with interspersed pauses. The first 50 calls per session were used for acoustic analysis. Calls distorted by background noises (mostly other barking dogs) or calls which had an overloaded recording level were discarded from the analysis.

Acoustic analysis

Signal analysis was performed on a PC using the signal processing software


53

HYPERSIGNALTM-Macro. Recordings were digitised at 16-bit quantization and 44-kHz sampling rate using a DSP32C PC System Board. Spectrographic analysis was completed using 512 points Fast Fourier Transformation, with 75% frame overlapping, a 30 kHz sampling frequency, and Hanning window. The result of this analysis is a spectrum representing the frequency distribution in the acoustic signal over a time segment of about 10 ms duration.

HNR calculation

For HNR calculation a 50 ms segment from the middle of the bark was cut out. The spectrum of this 50 ms segment was calculated via Fast Fourier Transformation (FFT). The spectrum curve was transformed to an EXCEL file. An EXCEL Macro was written to do the further steps. The noise level was estimated by a 10-point-moving-average from the original spectrum curve (figure 4.1). The largest distance between the harmonic level and the noise level was considered as HNR (Fig. 4.1). The HNR expresses no dimension since we used relative amplitudes. A flow chart of the procedure is given in figure 4.2.

Figure 4.1: Schematic drawing of an original spectrum curve with harmonic peaks (thin line) and the 10-point-moving-average curve of the same spectrum (thick line).


54

Figure 4.2: Flow chart of the HNR calculation procedure.

Testing the HNR calculation procedure with synthetic sounds

The graphical synthesiser of SASLAB-Avisoft (Specht 1998) allows the construction of harmonic sound elements stored as wave files. The number of harmonics, the amplitude of the single harmonics as well as the modulation of the fundamental can be chosen freely.

A sound file with 5 harmonics, with constant amplitude and constant fundamental frequency at 800 Hz was constructed. The amplitude was highest in the fundamental and systematically lowered by 5 dB, 10 dB, 15 dB and 20 dB, respectively, in the higher harmonics. The sampling frequency was at 22 kHz with a 512 points FFT.


55

The addition of noise occurred using a given noise file. A white noise file of a given average amplitude was used to construct ten noise files of different amplitudes by damping the original signal by 9 different dB filters using 6 dB steps. Each of the ten noise files was mixed with a copy of the initially constructed harmonic file resulting in ten files with different harmonic-to-noise-ratios.

Those ten synthetic sound files with different HNR were used to test the HNR calculation procedure, predicting a significant correlation between the results of the HNR calculation of the respective sound file and the file number according to the dB-filter step used for the noise file.

Bark evaluation by human listeners

The 20 dachshunds considered in this study were pairwise associated, and from 190 possible combinations 27 were chosen. Selection criteria were as following: The twenty dogs were sorted according their HNR values in three groups: six animals with the lowest HNR, six animals with the highest and seven with a middle HNR. Within each group three intragroup pairs were associated. Between the three groups three times six intergroup pairs, considering each group equally were associated.

This results altogether in 27 pairs (indicated in table 4.1). Each dog was considered at least ones, maximally in four pairs. About 15 successive barks from each dog were considered for one play back pair. Five persons were asked to judge the degree of hoarseness in pairwise playbacks. They were asked to decide which dog of the pair sounds more hoarse. They had to write down their decision on a spreadsheet. Prior to the 27 test pairs, the subjects were exercised with 4 training playbacks.

If the HNR measure relates to the human listening impression, a more than 50% concordance between the subject's results and the HNR scale should be expected.


56

Table 4.1: The pairs in the matrix with a frame represent those individuals which were used for the evaluation of the auditive impression by 5 subjects. The numbers in the framed boxes have to be read as ’the animal in the first column sounds more hoarse than the corresponding animal from the first row‘. The evaluation by the subjects corresponds with the HNR measure above the diagonal, but it does not below the diagonal. HNR means are given in the second row. n or c in the third row indicate if the individual belongs to the clinic or the normal sample.

 

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

HNR

7.1

9.7

9.9

10.6

11.7

11.8

12.9

13.2

13.6

13.7

13.8

14.0

14.2

14.3

14.4

14.9

15.5

16.3

17.6

17.8

N/c

c

c

n

c

n

n

c

c

c

n

n

n

n

n

c

n

n

c

c

c

1

-

 

5

 

 

 

5

 

 

 

 

 

 

 

5

 

 

 

 

 

2

 

-

 

3

 

 

 

 

3

 

 

 

 

 

 

 

4

 

 

 

3

 

 

-

 

1

 

 

 

 

 

 

 

3

 

 

 

 

 

 

 

4

 

2

 

-

 

 

 

 

2

 

 

 

 

2

 

 

 

 

 

 

5

 

 

4

 

-

 

 

2

 

 

 

 

 

 

3

 

 

 

 

 

6

 

 

 

 

 

-

 

 

 

 

 

 

3

 

 

 

4

5

 

 

7

 

 

 

 

 

 

-

 

 

 

 

 

 

 

 

 

 

5

5

 

8

 

 

 

 

3

 

 

-

 

 

 

5

 

 

 

 

 

 

 

 

9

 

2

 

3

 

 

 

 

-

5

 

 

 

 

 

3

 

 

 

 

10

 

 

 

 

 

 

 

 

 

-

 

 

 

 

 

 

 

 

 

5

11

 

 

 

 

 

 

 

 

 

 

-

 

 

 

 

 

 

 

 

5

12

 

 

 

 

 

 

 

 

 

 

 

-

 

 

 

 

 

5

 

 

13

 

 

2

 

 

2

 

 

 

 

 

 

-

 

 

 

 

 

5

 

14

 

 

 

3

 

 

 

 

 

 

 

 

 

-

 

 

 

 

 

 

15

 

 

 

 

2

 

 

 

 

 

 

 

 

 

-

 

 

 

 

 

16

 

1

 

 

 

 

 

 

2

 

 

 

 

 

 

-

1

 

 

5

17

 

 

 

 

 

1

 

 

 

 

 

 

 

 

 

4

-

 

 

 

18

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

-

4

 

19

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1

-

 

20

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

-


57


58

Multiparametric analysis

The multiparametric analysis of animal sound has been proven useful in a number of studies (reviewed in Schrader, Hammerschmidt 1997). The multiparametric analysis is based on the spectrogram of the signal. The present study was undertaken to verify if the classification of individuals as suggested by the HNR-measure could be reproduced with the multiparametric approach. In other words, is the information extracted by the HNR also obtained by the multiparametric approach?

The goal of this part of the study was basically to evaluate the HNR value by applying the multiparametric analysis.

For multiparametric analysis 1000 barks (50 barks from each of 20 individuals) were digitised using RTS 1.31 (Engineering Design, Belmont, MA). The digitised calls were transformed into frequency-time domain with Signal 2.29 31 (Engineering Design, Belmont, MA), using an FFT size of 1024 points, a sampling frequency of 40 kHz, a frequency resolution of 39 Hz, and time resolution of 5 ms. The resulting frequency-time spectra were analysed with LMA 8.0 (developed by Kurt Hammerschmidt). LMA is a software tool to extract different sets of call parameters from complex acoustic signals (Schrader and Hammerschmidt, 1997). 100 parameters were used in the analysis. These parameters describe the frequency structure, such as the fundamental frequency, and the distribution of the spectral energy, including the peak frequency and frequency range.

The ability of the parameters to separate individuals was tested first. The question was in which parameters is the intraindividual variance sufficient below the level of the interindividual variance, to differentiate individuals. This is necessary because parameters with individual variance not significantly beyond the total variance are not expected to be able to group individuals. In order to test for individual differences a stepwise discriminant function analysis was conducted. The discriminant function analysis provides a classification procedure that assigns each call either to its appropriate group (correct assignment) or to another group (incorrect assignment).

The correct assignment is a measure for the ability of the discriminant functions to predict a call's group membership. The prior probability of a correct assignment is equal for each group. The prior probability dependents from the group sample size insofar, as the larger the group's sample size the higher the certainty of the discriminant function.


59

It is robust against correlating variables. Variables that fail the tolerance test, that is, those that are almost a linear combination of the other independent variables are not permitted to enter the analysis. In the stepwise procedure the parameters are ordered according to their contribution to the correct assignment (identification of predictor variables).

In the second step we tested parameters for their ability to group individuals into HNR groups. Three HNR groups were made. The mean HNR ± standard deviation of the normal sample (n=10) was 13.2 ± 3.4. Considering this range of HNR = 9.8 to 16.6 three individuals are out of this range. For the sake of a sufficient minimum group size, cutoffs are set at 10 and 16, which results in three individuals with a mean lower than 10, three individuals with a mean higher than 16 and fourteen individuals between 10 and 16. Than a discriminant function analysis was applied to test how the calls are assigned to those three groups. To test the reliability of the discriminant function, first, the original data set was split into two. One half was used to establish a discriminant function (training set). The remaining data were than used separately in the assigning procedure (test set). Similar results in the assignment procedures (training versus test set) indicate that the discriminant functions based on the parameter set is appropriate to sort new calls into groups. Secondly, an individual's complete data set was used as test set, to test for the assignment of the calls of the 'new' individual ('leave one out' procedure).

4.3 Results

Illustrating the differences in the harmonic-to-noise-ratios, Figure 4.3 shows time series and spectrograms of three barks from three dogs. Harmonic structure can be recognized in the bark with the highest HNR while nearly no harmonic structure but only noise can be seen in the bark with the lowest HNR.


60

Figure 4.3: Spectrograms, time series and spectra of three barks with different HNR illustrating that in calls with high HNR values the harmonic peaks are stronger than in others.

Testing the HNR calculation procedure with synthetic sounds

Artificial sounds with a defined HNR were synthesised for testing the HNR calculation procedure. The correlation between the calculated HNR and the predicted HNR value was


61

r=0.95 (Spearman rank, n=10; P<0.001) suggesting the procedure is able to mirror at least the relative HNR relationships within a group of sound signals.

The HNR of twenty dogs

Figure 4.4 shows a diagram with the mean HNR ± standard deviation of twenty dogs considered in this study (the actual values are given in table 4.2). The mean of the normal sample is 13.2 ± 3.46, and that of the clinic sample is 13.5 ± 4.17. The samples differ significantly in their variances (F=1.45; P<0.001) but not in their medians (U=0.00013; P1=0.19; P2=0.38). This suggests that a ranking of animals according their HNR value is possible. The clinic and the normal sample is mostly overlapping. However some individuals from the clinic sample represent the far ends of the scale.

Figure 4.4: HNR mean and standard deviation of twenty individuals considered in this study.


62

Table 4.2: Actual values HNR mean and standard deviation of twenty individuals considered in this study. m - male, w - female

 

sex

HNR

normal sample

1

m

15.5 ± 2.3

2

w

9.9 ± 2.3

3

w

13.8 ± 3.2

4

m

14.2 ± 3.1

5

m

14.9 ± 2.9

6

w

13.7 ± 2.0

7

m

11.8 ± 2.4

8

w

11.7 ± 2.2

9

w

14.3 ± 2.3

10

w

14.0 ± 2.2

clinic sample

11

w

7.1 ± 1.4

12

w

17.8 ± 2.8

13

m

17.6 ± 3.3

14

w

16.3 ± 2.6

15

m

12.9 ± 2.9

16

w

13.2 ± 3.2

17

w

9.7 ± 2.8

18

m

14.4 ± 2.3

19

m

13.6 ± 2.7

20

m

10.6 ± 2.8

Comparison of the auditive impression of human subjects and the HNR

27 pairs of dogs were chosen and their bark sound were evaluated by human listeners to differentiate which of the two dogs sounds more hoarse. It was assumed that those dogs with the lower HNR sound more hoarse.


63

The correspondence between HNR and auditive impression in five subjects was 70, 75, 77, 78 and 80 % , i.e. 70 to 80% of the pairs were ordered by human subjects as predicted by the HNR (tab. 4.1). There were thirteen pairs ordered as predicted by all five subjects but there is no pair ordered wrong by all five. Three pairs were ordered as predicted by four subjects, six pairs were ordered as predicted by three subjects, three pairs were ordered as predicted by two subjects, and two pairs were ordered as predicted by only one subject. Two general conclusions can be drawn from the questionnaire. As expected, pairs laying far from each other on the HNR scale were easier to assess correctly than those very close to each other, and barks from the extreme ends of the HNR scale (hoarse or hyperkinetic) were obviously easier to recognise than those from the middle of the HNR scale. No pair was misclassified by all five subjects, but two pairs were misclassified by four subjects, three pairs were misordered by three subjects, six pairs were misordered by two subjects and three pairs were misordered by one subject.

Multiparametric analysis

Here several discriminant function analyses were run in order to assign calls to different groups. The groups represent the twenty individuals in the first analysis. The classification procedure yielded an average correct assignment of 94% to 20 individuals. The most decisive parameters for discrimination between groups were 'mean first dominant frequency band' and 'start frequency of the second quartile'.

For the second analysis, individuals were associated into three groups according their HNR value (figure 4.5 right diagram). This means 86% of the calls were ordered correctly into three groups, and the groups are based on the HNR measure (figure 4.5 left diagram). The first group contained three individuals with low HNR values, the second group contained fourteen individuals with middle HNR values and the third group contained three individuals with high HNR values. The second classification procedure yielded an average correct assignment of 86% to three HNR groups. A test with a 50% random sample confirmed the discriminant function: the average correct assignment of the training set was 88% and of the test set 81%. In figure 4.6 the correct and wrong assignment of calls to three HNR groups is shown separately for each individual.

In the third classification procedure single individuals were removed from the sample and the total individuals set were used as test set. The test with individuals from the second group (HNR between 10 and 16) revealed correct assignment of above 85 % in each of four tested


64

individuals. The average correct assignment of calls from either the first (HNR < 10) or the third (HNR > 16) HNR group was not or only sligthly above the chance level (33%) suggesting that the training with only two remaining individuals in those two groups is not sufficient.

Figure 4.5: The left diagram shows the distribution of 50 calls for each individual into three HNR groups according to its HNR value. The middle diagram shows the expected distribution of calls if the discrimination reaches a 100% correct assignment. The right diagram shows the result of the discriminant analysis, i.e. the assignment of calls into three groups.


65

Figure 4.6: Assignment of call to three HNR groups. In the first diagram 50 calls per individual were used for calculation the discriminant function and the same 50 calls were assigned into the groups. The second and third diagram shows the results when the sample was devided into a training and a test set. The size of the circles corresponds to the number of the calls.

4.4 Discussion

The harmonics-to-noise-ratio calculation procedure

In this study a simple and crude harmonics-to-noise-ratio (HNR) calculation procedure was applied to animal sounds. In human voice evaluation similar procedures are rated as crude. The goals in human voice therapy are different - it is mostly the restitution/creation of a 'normal' voice. The normal voice is defined: it is the voice, first, which is not conspicuous for the percipient in the average auditive impression, and second, it makes no discomfort for the speaker. In order to voice evaluation, experienced clinical human voice physicians primarily evaluate the auditive impression (e.g. Nawka 1994) and the laryngoscopic signs (e.g. Wendler, 1993), and yet than technical acoustic parameter play a certain role for voice evaluation (Seidner, pers. comm.).

In 1961 Lieberman proposed the first acoustic parameter in human pathological voice analysis. Since then, many different parameters have been introduced, which fall into three groups: parameters describing additive noise (turbulence) such as the harmonic-to-noise-


66

ratio (Yumoto et al. 1982; de Krom 1993) and parameters describing frequency modulation noise (jitter) and amplitude modulation noise (shimmer) (Kasuya et al. 1993) and parameters sensitive to additive noise and to modulation noise, like the glottal-to-noise-ratio (Michaelis et al. 1995). Methods most similar to those used in this study are described by Kitajima (1981). Kitajima (1981) has used a moving average method for the HNR estimation, which basically considers the ratio of the energy of the noise and the energy of the harmonics. The level of noise is estimated by calculation of the moving average over the spectrum curve. Kasuya et al. (1986) introduced the normalised-noise-energy (NNR). It represents the ratio between the energy of noise and total energy of the signal. In both methods, the noise energy is directly obtained from the spectrum. In the NNR procedure, the noise energy is assumed to be the mean value of both adjacent minima in the spectrum.

The problem in both methods, as in other procedures (like the cepstrum based) is due to jitter and shimmer. While increasing the modulation noise (jitter and shimmer) the harmonics are broadened and the minima of the spectrum are less deep. As a consequence noise energy is overestimated.

However the dog bark is a short amplitude and frequency modulated utterance. This means that frequency peaks become broader if considered over large time segments. The differentiation into additive noise and modulation noise seems not useful, because there are only few data about the laryngeal behaviour during barking (Solomon et al. 1992). To overcome this problem only a short time segment was considered, assuming the fundamental frequency modulation is only weak. Visual inspection of a considerable number of those spectra confirmed this aspect.

Another problem of the moving average procedure are vocal tract effects. All those effects imaginable (e.g. formant tuning) should be minimised (systematised) by considering only one dog breed with a small range of body mass (and body size and thus vocal tract length).

Comparison of methods

For a validation of the HNR-measure, first the algorithm has been tested with artificial sound signals and second the dog barks have been undertaken an auditive evaluation by humans and the multiparametric analysis was applied. The validation of the HNR measure was necessary, first because the algorithm used for calculation is very crude, and second because there are no data for comparison in animal bioacoustics.

In a first approach, synthetic sound signals with a defined HNR to test the procedure were


67

used. The synthetic sounds could be differentiated by our algorithm as predicted indicating that the algorithm was working safely.

The aim of the multiparametric approach was to reproduce the ordering of the dogs according to the HNR scale. The multiparametric approach has already been proven useful in bioacoustics. The results of a first discriminant analysis showed that a subset of parameters is good for individual identification and following is fulfilling the demand for a large interindividual and a low intraindividual variance. Prior the next discriminant analysis, the dogs were assigned into three groups because the philosophy of the discriminant analysis is to assign elements (calls) into given groups optimally. The subset of parameters allowed a 85 % average correct assignment to the three groups. However (in the leave one out procedure) the poor correct assignment of a new individual to either the first (HNR <10) or third HNR group (HNR >16), was probably due to insufficient low number of individuals (only two) considered in the training set. This hypothesis was supported by the very good high percentage of correct assignment to the middle HNR group (HNR between 10 and 16).

It was also shown that there is a relationship between the HNR ranking of the individual and the auditive impression of its barks assessed by five subjects. Kojima et al. 1980 have undertaken a similar approach for the human voice. They used also the HNR measure for ranking a sample of voices, and human subjects sorted this sample according to their impression of hoarseness. The concordance between the HNR ranking and the subject's ranking was very high. Most of the dogs are within the range of mean ± standard deviation of the normal sample.

Possible physiological reasons for the observed dysphonia

Three individuals were below and three above the range of HNR 10 to 16. The animal with the lowest HNR belongs to the clinic sample. It was housed in the clinic for 18 days. During that time the animal was frequently vocalising (barking). We assume that the observed hoarseness (and lowerence of the HNR) was due to an exhaustion of the vocal folds (excessive barking/hyperphonation) which means microlesions of the mucosal cover impacting the vocal fold vibration behaviour. This phenomenon is well known in humans (Wendler et al. 1996) and it has also been suggested to be the reason for a cat's dysphonia (Riede, Stolle-Malorny 1998, 1999). In vitro studies showed that hyperphonation can affect a dog's vocal fold tissue (Gray et al. 1987; Gray, Titze 1988). Similar effects to the human vocal folds would have massive impacts on the voice (e.g. Fex, Fex, 1986).

Three dogs, with the highest HNR values belong to the clinic sample. They suffered from


68

intervertebral disc extrusion, a common disease of the dachshund breed (Grevel, Schwartau 1997 a,b). Intervertebral disc extrusion occurred in all three dogs in the transition from the thoracic to the abdominal backbone. The extrusion results in a compression of the spinal cord which adversely affects the normal nerval supply behind (caudally) the affected point. Neurologic examination revealed paresis of the hind legs with intact spinal reflexes in all three dogs. After detailed diagnostic investigations the aim of the therapy was to decompress the spinal cord. This was done by performing a hemilaminectomy in all three dogs. Laryngeal inspection by direct visualisation prior and after the operation (and tracheal intubation) showed a normal picture. The dysphonia occurred one day after the surgical decompression and was reversible toward normal sounding barks within three days. The dysphonia in those cases can be considered as a functional one since the laryngoscopic picture showed no obvious injuries or harms. In humans, the functional-hypertensive dysphonia is characterised by an increased HNR (Wendler et el. 1996), and increased harmonic energy suggests increased HNR as in the three cases here. In functional-hypertensive dysphonia, the laryngeal stage of voice generation is characterised by an increased tension of the vocal fold muscle and optional by an increased subglottic pressure (Wendler et al. 1996).

Six other dogs with thoraco-lumbal intervertebral disc extrusion were considered in this study. They did not express extrem HNR values. Three of them had undergone surgical decompression of the spinal cord by hemilaminectomy, and three were treated with antiphlogistic therapy. Obviously the dysphonia is not a regular occurring symptom, it follows only in some cases after surgery (hemilaminectomy).

One explanation for a hypertension in the laryngeal configuration in the hemilaminectomized dogs is pain. However, we observed no disappearance of the hypertensive sound (i.e back to a normal HNR) under the regular pain reducing butyrophenon (Temgesic?) therapy.

A second plausible hypothesis explaining the observed hypertensive sound is a peripheral nerval stimulation, caused by the surgery, which resulted in a retrograde stimulus transport via central areas to the vagus nerve. Such a peripheral stimulation and the retrograde stimulus transport have already been observed for the thermoregulative system (Romanowsky et al. 1997; Szekely et al. 1997). However, further research is necessary to uncover the relationship between this hypertensive dysphonia and the disease.

HNR variability and communicative relevance of the dog bark

Some authors discuss a communicative relevance according to the different harmonic-to-noise-ratio, since they observed different acoustic structure of the dog bark in different contexts (Tembrock 1976; Feddersen-Petersen 1996; Fox 1978). In this study only one


69

context of vocalization was considered.

In general, the communicative relevance of the dog bark is discussed controversely. While Coppinger, Feinstein (1990) degrade the dog bark to a non-communicative utterance, Bleicher (1963), Tembrock (1976), Althaus (1982) assume that barking is an expression of a general arousal of varying intensity. Feddersen (1978), Feddersen-Petersen (1996), Fox (1978) hypothize that barking sounds carry particular information. Feddersen-Petersen (1996) relates her arguments for different meanings to certain aspects of the acoustic structure. Barks in different contexts show different degrees of tonality what refers to a varying ratio of the harmonic energy and the noise energy (Feddersen-Petersen 1996). Since the communicative relevance of dog barking is still hypothetical in many respects, the application of HNR for a parametric analysis holds considerable promise for future research.


[Front page] [Acknowledgements] [Attached Audiofiles] [1] [2] [3] [4] [5] [6] [7] [Bibliography] [Declaration]

© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.

DiML DTD Version 2.0
Zertifizierter Dokumentenserver
der Humboldt-Universität zu Berlin
HTML - Version erstellt am:
Thu Sep 28 14:01:51 2000