This was a prospective comparative study wherein normal voices and abnormal voices were collected at Department of ENT, BLDE (Deemed to be University) Shri BM Patil Medical College, Hospital and Research Centre, Vijaypur, Karnataka, India. The study was conducted from March 2016 to Feb 2017. Ethical clearance was taken from the Institutional Ethical Clearance Committee (ECR/383/INST/KA/2013/RR-16). After getting informed consent from each person, they underwent thorough clinical evaluation by ENT surgeon.

With anticipated least difference of means of study variable between the cases and control groups as 0.0185 and anticipated SD (standard deviation) as 0.0236, with 90% power and 5% level of significance, the minimum sample of 43 under each group was calculated.

A total of 43 normal voices were collected from the database of young healthy adults between age of 18 to 28 years, of which 23 were males and 20 were females. Any person having history of smoking or any other pathological conditions which made them unfit for the study were excluded.

Forty-three patients (23 males and 20 females) between 18 to 55 years presenting with hoarseness and lesions of larynx like vocal nodules, vocal polyps, chronic laryngitis, voice abuse, early glottis carcinoma etc., were selected during the study.

Voices were recorded in a sound treated chamber using a unidirectional microphone (Sony Audio technical 250*L). The sustained /a/ vowel signal was recorded for three seconds, using PRAAT version 5.4.04 [8]. The intensity was controlled using VU meter built into PRAAT. From the recorded sample voice of three seconds duration, sampling frequency of 44100 Hz, spectrograph was extracted and the more stable and uniform one second was selected and saved as a sound file. A PRAAT script was written which can batch process all files into a folder to extract four parameters, Jitter (ddp), Shimmer (dda), Harmonic Noise Ratio (HNR) and Median pitch.

These parameters were analysed for different variations like Mean, standard error of mean and Kolmogorov-Smirnov ZH values. Median Pitch, Jitter, Shimmer, and HNR between normal and abnormal voice groups of males and females were compared. The level of significance was set at 5% (p<0.05) with SPSS Version 23.

The values of Mean, standard error of mean, and Kolmogorov-Smirnov ZH values of Median Pitch, Jitter, Shimmer, and HNR between normal and abnormal voice groups of males are given. The Jitter had a p-value of 0.026 which was significant but Median pitch, HNR and shimmer did not show the significance [Table/Fig-1].

When the Mean, standard error of mean, and Kolmogorov-Smirnov ZH values of Median Pitch, Jitter, Shimmer, and HNR of females were compared between normal and abnormal voice groups, the shimmer had a p-value of 0.035 which was significant but Jitter, Median pitch and HNR did not show the significance [Table/Fig-2].

Voice is a multidimensional measure which requires various methods for its assessment like Perceptional measures, acoustic analysis, aerodynamic measures, video-stroboscopy, non-linear dynamic measures etc. The objective of the present study was to compare voice parameters of abnormal voices with normal voices. This study also shows that with the availability of freely downloadable software like PRAAT, ENT surgeons and Speech pathologists can easily use the software to analyse the voice effectively.

As early as 1987, Baken RJ had given details about few valid techniques for acoustic assessment of vocal dysfuction which are easily accomplished with current instrumentation [9]. Leiberman P recorded 23 voices from the speakers who had pathologic growths on their vocal cords. It was found that they had larger perturbations than did normal speakers with the same median fundamental periods. This may be related to size of pathological growth [10]. This showed that study of perturbation measures is important for assessment of voice. Hence voice parameters like Jitter (Leiberman P), shimmer (Horii Y), and harmonics-to-noise ratio (Yumoto E et al.,) were extensively studied [10-12]. A study by Hillman RE et al., demonstrated that acoustic measures alone could be highly accurate in determining the presence/absence of a voice disorder [13]. According to Eskenazi L et al., the two most useful parameters for predicting vocal quality were the Pitch Amplitude (PA) and the HNR [14].

Many authors studied both normal and pathological voices using the parameters like jitter, shimmer, and HNR. Di Nicola V et al., studied 208 subjects (148 with dysphonia and 60 normal) using computerised digital sonography [15]. HNR developed by Yumoto E et al., was analysed and found to be highly sensitive as values were different in dysphonic patients. The comparison between the average HNR recorded in those patients (1.697 dB) is significantly different from that recorded in the normal subjects (11.169 dB) (p<0.001 [12].

A similar study was done for voice analysis by Bielamowicz S et al., using C Speech, Computerized Speech Laboratory, SoundScope, and a hand marking voice analysis system. Sustained vowels from 29 male and 21 female speakers with mild to severe dysphonia were used. They felt that measures of perturbation in the various analysis packages use different algorithms, provide results in different units, and often yield values for voices that violate the assumption of quasi-periodicity. As a result, poor rank order correlations between programs using similar measures of perturbation were noted [16]. Their sample size is almost same as this study.

In this study Mean, standard error of mean, and Kolmogorov-Smirnov ZH values of Median Pitch, Jitter, Shimmer, and HNR were compared between normal and abnormal voice groups.

When the above parameters were compared among normal and abnormal male voices, the value of Jitter (ddp) was significant (p-value was 0.026, i.e., p<0.05). Median pitch, Shimmer (dda) and HNR values were different but not significant. (p-values were 0.649, 0.414 and 0.124 respectively).

When these parameters were compared among female normal and abnormal voices, shimmer values were significant. (p-value was 0.035 i.e., p<0.05).

Limitations are present for acoustic analysis when it comes to assess aperiodic voices. Titze IR discusses three types of voices depending on the periodicity of the voices. While types 1 and 2 have relatively better periodicity, the type 3 voices have aperiodic waves [17]. Type 3 voices pose problems on correctly analysing the voice parameters. In fact recently, the study by Núñez Batalla F et al., showed that there is no difference between using PRAAT and Dr. Speech for analysis of type 3 voices. Even PRAAT can analyse effectively type 3 voices [18].

Many authors felt other methods like nonlinear dynamic measures are more indicative of differentiation in such cases particularly in chaotic voices. Jacqueline BF et al., studied voices of Indian population comprising adults and elderly. They concluded that parameters like correlation dimension D2 (a voice measure of nonlinear dynamic analysis) are better assessors. The anatomical alterations in the vocal mechanism that occur for any pathological conditions result in higher values of correlation dimension. Thus, it can be considered as a useful tool in the assessment of voice. However, they also felt that the existing voice analysis techniques available to the voice clinician cannot be replaced but nonlinear measures can be added to the existing battery of tests [19].

In this study, comparison of the recorded abnormal voice parameters were done with normal voices and added to a voice database. This normal database is continuously added with voices recorded at our institution which naturally reflects the local sample population effectively. The same authors also proposed an institutional voice database, standard voice analysis tool, and method of voice measurement [7].

According to Stemple JC et al., normal database should report demographic details of their local sample populations to account for factors that influence the instrumental results such as gender, age, health history and local database [20]. Recording techniques and sample tasks may vary across the studies as well as equipment and analysis routines [21]. These factors limit the ability to compare different findings. A practical solution is to collect local norms by measuring a large group of normal speakers as a separate sample.

Di Nicola V et al., on comparison of normal and abnormal voices, felt that forensic evaluation of dysphonia needs application of strict, precise, correct sampling and analysis method following well-defined rules. Further comparison of normal and abnormal voices, HNR was analysed and found to be highly sensitive [15].

Jiang et al., examined objective acoustic analysis methods nonlinear dynamic and traditional perturbation measures like jitter and shimmer, to assess voices of patients with vocal nodules and polyps [22]. The jitter or shimmer showed no significant changes but correlation dimension, a parameter of nonlinear dynamic measure of voice showed significance. They concluded that the combination of traditional perturbation and nonlinear dynamic measures may improve our ability to provide objective clinical analysis of voices with vocal mass lesions.

The cost of these analysis programs is undoubtedly high and routine use by clinicians is not possible. However, development of free computer application with widest diffusion like PRAAT has greater capabilities of analysing acoustic signals [18]. The authors also recommend PRAAT program as valid, reliable, and easily manageable and has minimum equipment requirements.

More and more normal and abnormal voices from local area are to be added to institutional voice database. This helps in overcoming bias and adds to further research on voice assessment.

The age range of normal voice study cases were 18-28 years whereas that of abnormal voice were between 18-55 years. This happened because abnormal voices were more common in older age group. The sample size is small. Hence, age difference was not taken into consideration. Though the collected sample size was 63 abnormal voices (32 males and 31 females) only 43 voice files could be assessed by the program. Remaining 20 voice files became corrupted and were not analysed by PRAAT Program. The future studies are to be done on a relatively larger number of subjects from a variety of dysphonia population so that changes in parameters in each specific pathology can be defined.

Acoustic voice analysis is still a valuable technique which enables voice clinicians to compare voices to differentiate them into normal and abnormal. But this requires a robust normal database of the local and regional demographic voice samples as the recording techniques, equipments, age, sex and analysis programs differ from one another. This method can provide a non-invasive and objective tool to identify and document abnormal voices.