Acoustic Analysis of Voice Signal of Patients with Unilateral Laryngeal Paralysis a View to Objective Evaluation after Rehabilitation

The evaluation of the voice quality of laryngeal paralysis after rehabilitation in Oto-Rhino-Laryngologist (ENT) services is based on a subjective method using perceptual judgement. In this method the orthophonist is responsible for evaluating the quality of the voice by listening, which results in an unreliable perceptive assessment. From there, a reliable perceptive evaluation requires several expert speech-language pathologists and several listening sessions, which increases time and human resources, and does not allow regular use in medical practice [1, 2]. Another major disadvantage of subjective evaluation is the inter- and intra-listener variability in the perception of the voice by a board of experts. This variability can be influenced by context, emotional state or listener’s attention [3, 4]. The objective evaluation of the voice based on the acoustic analysis makes it possible to make measurements, obtain indices, identify trends, and do so quantitatively and objectively [5]. This instrumental method can assist the speech language pathologist in assessing the quality of the voice during the rehabilitation period [6-8].

In this sense, research has been carried out to objectively assess the voice in the Algerian clinical setting, nevertheless, they remain insufficient in relation to the protocol and conditions of recording of pathological voice which require technical precautions to acoustic evaluation [1, 9].

In this study, we propose an acoustic analysis for an objective evaluation of pathological voices in subjects with laryngeal paralysis at the ENT department of the Bab El Oued Hospital. Algiers (Algeria). The objective of this work is to show that the acoustic analysis and its results can help the othophonist in rehabilitation, and to evaluate, objectively, the evolution of this rehabilitation over time. We will show in this work, the degree of convergence of acoustic parameters of pathological voice and normal voice after rehabilitation, we will also mount the contribution of the objective method based on the acoustic analysis of the vocal signal on the evaluation of the voice after rehabilitation at the level of the ENT service.

3.1 Selected subjects

The subjects selected for this works consists of nine Algerian female patients aged 42 to 56 with Unilateral Laryngeal Paralysis (ULP), six are left and three are right (Table 1). Patients over the age of 56 were eliminated in this study to give reliability to our results (Figure 2). A recording is made after a 9-month speech rehabilitation. In this study, we only selected patients who followed a regular rehabilitation protocol. The same corpus was pronounced by 3 normal female speakers between the ages of 40 and 50 years, not presenting voice disorders (reference standard).

Table 1. Pathological population

2.png

Figure 2. Distribution of patients by age

3.2 Recording equipment’s

The voice corpus was recorded with an external M-audio pro sound card, with a Signal/Noise ratio of 100 dB and 24 bits of resolution. We selected a sampling frequency of 44.1 kHz. A dynamic microphone of the Sennheiser e815S type is used for recording with sound software Sound Forge version 10. To measure the intensity, we used a digital sonometer model Chauvin Arnoux C.A 832, has a dynamic sound level 37 to 130 dB. We calibrated the sonometer on the A-weighting that approximates the sensitivity of the human ear. The A-weighting allows to measure sound levels between 37 and 85 dB. The Distance between a patient’s mouth and a sonometer was precisely respected for all patients.

3.3 Protocol recording and condition

The voice recordings were made in an acoustically quiet room to eliminate parasitic sound sources. When recording the vocal corpus, a distance of 5 cm is respected between the microphone and the patient's mouth. The microphone is placed at 45° laterally to the mouth, its gain has been adjusted to have an optimal quality of the recording and to avoid the saturation of the sound. The patient should hold the vowel [a] as long as possible with a strong inspiration at comfortable intensity [12]. No demonstration is given to the patient beforehand, not to influence the height.

3.4 Acoustic analysis

The acoustic data studied were extracted using PRAAT software version 6.1. [13]. A duration of two seconds was chosen in the most stable part of the recorded signal. Since dysphonia is primarily concerned with the voice source, we propose in this article an acoustic analysis based on measurements of the instability of the frequency and amplitude of the laryngeal vibration [1, 2]. In this study, the acoustic parameters chosen are: the average fundamental frequency (Pitch) Fo, the Coefficient of Variation of Fo (CV) which measures the overall instability of the voice over the whole sample, the jitter which evaluates instability of fundamental frequency, the intensity (energy), the shimmer which evaluates instability of the amplitude of Fo, the energy ratio of the Harmonic spectrum and that of the Noise spectrum (HNR) and the Maximum Phonation Time (MTP) [14, 15].

The acoustic parameters used in this study were defined and calculated as follows:

The mean fundamental frequency Fo is a very important acoustic indicator for determining the biomechanical characteristics of the vocal cords. To measure it, PRAAT has a frequency interval of values of 75 to 600 Hz [1]. The average frequency can be calculated by the following relationship:

Fo average $(H z)=\frac{1}{N} \sum_{i=1}^{N} F 0_{i}$    (1)

Jitter factor is the most commonly used parameter in vocal pathology. It is available since version 4.0 of PRAAT [10]. It consists of averaging all the differences, in absolute value, between two consecutive periods of the signal, and dividing this mean by the average length of the periods of the signal [1, 16, 17].

Jitter factor $=\frac{\sum_{i=1}^{n-1} \frac{\left|T_{i}-T_{i+1}\right|}{n-1}}{\sum_{i=1}^{n} \frac{T_{i}}{n}}$    (2)

The shimmer factor is based on exactly the same principle as the Jitter factor. The mean of the absolute differences between the maximum amplitude of two successive periods is divided by the mean of the maximum amplitudes of each period [1, 16, 17].

Shimmer factor $=\frac{\sum_{i=1}^{n-1} \frac{\left|A_{i}-A_{i+1}\right|}{n-1}}{\sum_{i=1}^{n} \frac{A_{i}}{n}}$    (3)

The Signal to Noise ratio HNR is an acoustic parameter used to indicate the rate of presence (or absence) of noise during the phonation. It is calculated by a temporal method [18]. Twenty-five consecutive cycles of the voiced signal allow to establish an average undulatory form; the subtraction between the original signal and the average form allows to obtain a residue considered as the noise. If H is the energy of the periodic averaged wave and N is the energy of the residual noise, the H/N ratio calculated regularly, provides the value of the HNR possibly converted into dB.

The TMP varies according to lung capacity, and according to glottic leakage during phonation. A reduction in TMP may be the result of low lung capacity or glottic air loss. TMP can be measured by a stopwatch but for reasons ease, we measured it directly in the time window of PRAAT.

The aim of this work was to evaluate the results of the vocal rehabilitation of a unilateral laryngeal paralysis type dysphonia with an objective evaluation, based on physical parameters. After 9 months of rehabilitation, overall, the analysis of the results obtained shows a general stability of the laryngeal vibration and therefore the absence of problems of pneumo-phonic control. This overall finding is confirmed by the subjective evaluation by the orthophonist. however, the study of the results obtained by acoustic analysis during the rehabilitation phase, allowed us to extract some important findings. We noted a significant improvement in the average value of Pitch Fo with its coefficient of variation, which explains the normal functioning of the motionless vocal cord due to recurrent nerve paralysis. However, we found a very low improvement in Jitter in the first six months of rehabilitation, which shows short-term instability of the laryngeal vibrator during this period of rehabilitation. This can be explained by the patient’s slow adaptation with the rehabilitation method adopted by the ENT department of the Bab El Oued Hospital which uses the François Le Huche method as a rehabilitation technique [8]. The low results for intensity, the HNR and MTP compared to the reference standard in the first phase of rehabilitation (after six months) show a significant glottic leakage caused by a bad adduction of the vocal cords. These parameters were improved at the end of rehabilitation. On the other hand, we recorded an increase in the Coefficient of Variation and the Shimmer at the end of rehabilitation, which raises questions about this increase in values of these two parameters. These results still show the presence of short-term instability in the amplitude of the vibration of the vocal cords at the end of rehabilitation, despite the improvement of the intensity. This is explained by a rehabilitation based on breathing work more than vocalization work or sustained vowels which makes it possible to vibrate the vocal cords in good condition [8].

Overall, we found a strong correlation between objective evaluation based on acoustic parameters and subjective evaluation. However, we noted insufficient values for certain acoustic parameters, he latter is confirmed by the orthophonist, who affirm that the majority of patients refuse to apply the entire rehabilitation protocol as singing exercises, this gives a low stability of the laryngeal vibration [19].