Spectral and Temporal Characteristics of Vowels in Konkani

Abstract

The present study was undertaken to study the acoustic characteristics of vowels using spectrographic analysis in Mangalorean Catholic Konkani dialect of Konkani spoken in Mangalore, Karnataka, India. Recordings were done using CVC words in 11 males and 19 females between the age range of 18–55 years. The CVC words consisted of combinations of vowels such as (/i, i:, e, ɵ, ə, u, o, ɐ, ӓ, ɔ/) and consonants such as (/m, k, w, s, ʅ, h, l, r, p, ʤ, g, n, Ɵ, ṭ, ḷ, b, dh/). Recordings were done in a sound-treated room using PRAAT software and spectrographic analysis was done and spectral and temporal characteristics such as fundamental frequency (F0), formants (F1, F2, F3) and vowel duration. The results showed that higher fundamental frequency values were observed for short, high and back vowels. Higher F1 values were noted for open vowels and F2 was higher for front vowels. Long vowels had longer duration compared to short vowels and females had longer vowel duration compared to males. The acoustic information in terms of spectral and temporal cues helps in better understanding the production and perception of languages and dialects.

Introduction

Speech sounds comprise of vowels and consonant sounds. Vowels are the class of sounds that are produced with unobstructed and open vocal tract above the level of larynx i.e. no strictures in the vocal tract [1]. Vowels retain the same qualities irrespective of the pitch in which they are produced [2]. Vowels are usually described with respect to their fundamental frequency, formants, formant patterns, amplitude, bandwidth, and duration. Among these characteristics, it is assumed that fundamental frequency, formant pattern and vowel duration play an important role in the perception of vowels [3]. Further, the vowel duration and formants play a crucial role for identification of individual phonemes in an individual’s speech. Although the vowel duration plays no role in identification, it plays an important role in helping the listeners to differentiate between similar vowels. Though vowel duration does not help in identification of any of the vowels, it plays a major role in helping the listeners to differentiate the vowels that are similar in their spectrum [4]. Though vowels contain a large number of formants during their production in the vocal tract, for practical purpose, only the lower three or four formants are studied.

The vowels of different languages may sound the same, but there are subtle variations in them between languages and dialects [5]. There have been attempts made to investigate the acoustic characteristics of vowels in different languages and dialects across the world [1, 6, 7]. Studies have also been carried out on underexplored languages [7–9].

Kpodo [8] studied 660 vowel tokens from Akpafu and Lolobi dialects of Siwu language and noted that Lolobi vowels are higher than the Akpafu vowels, except for/I/which is at the same position, but more back in Akpafu. Vowels in Austronesian languages spoken in Ireland such as Sursurunga, Notsi, and Tiang were studied for formant duration and formant frequencies. The findings revealed that in Sursurunga, the mid-open central vowel is more back than open central vowels in Sursurunga. Further, findings were supportive of the notion that Tiang has three distinct central vowels [7]. Savithri et al. [9] analyzed the effect of stops on glottal and pharyngeal stops on vowel characteristics in Nuu-chah-nulth (Nootka), a Wakashan language spoken on the West coast of Vancouver Island. They reported formant values to be significantly different for vowels adjacent to the plain glottal stop as compared to the pharyngealized glottal stop.

Vowels in 13 Indian languages were compared for the base of articulation differences among 10 native speakers. Significant differences were noted across languages, vowels, and genders [10]. Basu et al. [11] carried out the acoustic analysis of vowels in three languages (Nagamese, Ao and Lotha) spoken in Nagaland to explore the vowel duration and formant characteristics. Except for Nagamese, there was an improvement in recognition of vowels upon introducing the third formant. Yerraguntla [12] investigated acoustic characteristics of vowels in Telugu language among children and adults and reported changes in acoustic characteristics with respect to age, gender, regions and consonant contexts. Maxwell and Fletcher [13] reported of differences in the vowel system of Indian English as compared to native speakers which are dependent upon their first language (L1).

Konkani is one of the 22 scheduled languages mentioned in the eighth schedule of the Indian Constitution and the official language of the Indian state of Goa. It is a minority language in Karnataka, Maharashtra, Kerala, Dadra and Nagar Haveli, and Daman and Diu. It belongs to the southern group of the Indo-Aryan languages, is spoken in Goa, Karnataka, Kerala and Maharashtra in India. The dialects of Konkani are dependent upon the religion, region, caste, and influence of local language. There are various dialects of Konkani spoken in India–Northern Konkani spoken in Maharashtra, such as Malvani, that is culturally tied with Marathi; Central Konkani in North Karnataka and Goa, where Konkani is tied with Portuguese and Kannada; Southern Konkani which is spoken in the Mangalore and Udupi, in close contact with Tulu and Kannada. In Mangalore, the Catholic community speaks a dialect of Konkani called as the Mangalorean Catholic Konkani or Mangalorean.

As per the eighth schedule of the Indian Constitution, there are 22 official languages in India. The Census survey [14] identified over 121 languages (22 official/scheduled and 99 unscheduled) and over 19,500 dialects and mother tong. It would be interesting to understand the subtle differences existing between them the different languages from the linguistic and acoustic perspective. Konkani was added as an official language in the 8th schedule 71st Amendment of the Indian Constitution. Konkani despite having a small population in Karnataka shows a high number of dialects based on religion, region, caste and influence of local language. Studies have pointed out towards the differences in quality among the vowels that might have the same transcription using similar symbols across languages. Thus, there is a need to study and quantify these differences for vowels used in different languages [8, 15]. Further, there is a dearth of studies describing the acoustic characteristics of vowels in Mangalorean Catholic Konkani dialect of Konkani. The present study was undertaken to study the acoustic characteristics of vowels using spectrographic analysis in Mangalorean Catholic Konkani dialect of Konkani spoken in Mangalore, Karnataka.

Method

A Cross-section study design was used to study the spectral and temporal characteristics of vowels across consonant contexts in Mangalorean Catholic Konkani dialect of Konkani spoken in Mangalore, Karnataka, India. The study was approved by the Institutional Ethical Committee.

Participants

Thirty participants in age ranges of 18–55 years were picked randomly and included. The inclusion criteria was native speakers of Mangalorean Catholic Konkani dialect of Konkani spoken in Mangalore, Karnataka in the age range of 18–55 years. All participants had hearing sensitivity within normal limits as well as age-appropriate speech language skills as tested by speech-language pathologist. All participants were rated to have normal speech intelligibility. Any individual with any structural deformity affecting their normal speech production was excluded.

Material

A list of 50 meaningful monosyllabic words consisting of CVC syllables was selected from the magazines of the dialects of Konkani (Magazines—Rakno and Sevak) (“Rakno,” n.d.; “Sevak,” n.d.). Five native speakers of the dialect were asked to do a familiarity rating using a three-point rating scale comprising of very familiar, familiar and unfamiliar. Words rated as familiar and very familiar were included in the final word list. The final word list comprised of 30 words that contained both short and long vowels present in Konkani language in different consonant contexts.

The vowels, consonants and their frequency of occurrence in different contexts are given in Tables 1, 2 and 3 respectively. A carrier phrase of “/ho sobdh (target word)/” (This word is _____) was used to ensure uniform stress and intonation pattern across all the words [12]. The target word was kept in the final position of the carrier phrase. A PowerPoint presentation was made with each word on a separate slide. The slide background was plain white and the text appeared in black colour.

Table 1.

List of Konkani vowels included in the study

	Front	Central	Back
High	/i/, /i:/	/ɵ/	/u/
Mid	/e̞ /	/ə/	/o̞/
Low		/ɐ//ӓ/	/ɔ/

Table 2.

List of Konkani consonants included in the present study

Consonant included	Description of the consonant
/m/	Bilabial nasal sound
/k/	Voiceless unaspirated velar plosive
/w/	Bilabial semi-vowel
/s/	Voiceless alveolar fricative
/ʅ/	Voiceless alveo-palatal fricative
/h/	Voiceless glottal fricative
/l/	Alveolar lateral
/r/	Alveolar trill
/p/	Voiceless bilabial plosive
/ʤ/	Voiced unaspirated palate alveolar affricate
/g/	Voiced unaspirated velar plosive
/n/	Alveolar nasal
/Ɵ/	Voiceless interdental fricative
/ṭ/	Voiceless alveolar stop
/ḷ/	Retroflex
/b/	Alveolar lateral
/dh/	Voiced interdental fricative

Table 3.

Frequency of occurrence of vowels in different contexts

Vowels	Consonants
	/m/	/k/	/w/	/s/	/ʅ/	/h/	/l/	/r/	/p/	/ʤ/	/g/	/n/	/Ɵ/	/ṭ/	/ḷ/	/b/	/dh/
/i/	1				1								1	1
/i:/										2						1
/Ɵ/		1													1
/e̞ /			1							1
/ə/		1						1
/u/				1		1						2
/o/			2	1								1
/Ͻ/	1	1					1		1			1		2		1
/ɐ/		1	1	1		1						1	2				1
/ӓ/	2	3	2	2			4	1			1	5	1			1

Procedure

An informed written consent was taken from each participant after explaining him about the study. The recording was done in a sound-treated room. The recording was done using a condenser microphone with the setting at Cardioid pattern held at a distance of 7–10 cm at 45° angle from the speaker’s mouth. The recordings were carried out using a laptop (Hewlett Packard, core i5 7th Generation) using PRAAT software using a sampling rate of 44,100 Hz. The words were presented using PowerPoint presentation one at a time. The participants were instructed to read the target word as a part of the carrier phrase. All the recordings were saved as.*wav format for analysis. Nine hundred tokens (i.e. 30 participants * 30 words) served as the sample used for further analysis. The recorded stimulus was judged by a native Konkani speaker speech-language pathologist to check for clarity, pronunciation and rule out discrepancy.

The spectrographic analysis of the tokens was done using PRAAT software. The spectral and temporal characteristics of vowels such as fundamental frequency (F0), formants (F1, F2, F3) and vowel duration were extracted. Each recorded token was subjected to spectrographic analysis to identify the spectral and temporal characteristics. All the target words comprised of CVC combination true words and only the V (vowel) portion was analyzed. The target vowel was selected using the two cursors on the spectrogram. The average fundamental frequency and formants (F1, F2, F3) was obtained and entered in Excel sheet for further analysis The difference in the starting and ending positions of the cursors of the target vowel served as the vowel duration (in seconds) which was converted to milliseconds for further analysis. In order to ascertain interjudge reliability, 20% of randomly selected token were re-analyzed by one of the other investigators and noted to have good inter-rater reliability.

Descriptive statistics was done to summarize the continuous variables. The inter-judge reliability was assessed using paired t-test. SPSS 17 was used for carrying out the statistical analysis. XLSTAT Microsoft Office Excel plug-in was used for generating the scatter plot with ellipses fit.

Results

Thirty native Mangalorean Catholic Konkani dialect speakers of Konkani spoken in Mangalore, Karnataka were selected for the present study. The mean age of the speakers was 30.77 (± 10.56) years, range 19–54 years and comprised of 11 males and 19 females. The spectral characteristics such as fundamental frequency (F0), first formant (F1), second formant (F2) and third formant (F3) as well as the temporal characteristic of vowel duration (in m sec) were considered for the present study.

Spectral Characteristics of Vowels

The findings for the spectral characteristics of vowels such as fundamental frequency (F0), first formant (F1), second formant (F2) and third formant (F3) in Hz are as depicted in Table 4.

Table 4.

Mean values of the fundamental frequency (F0), first formant (F1), second formant (F2) and third formant (F3)

		/i/	/i:/	/e̞/	/ɵ/	/ə/	/ɐ/	/ä/	/u/	/o̞/	/ɔ/
F0	M	162.61	153.09	143.35	154.17	153.59	152.83	146.96	163.63	147.72	146.28
	W	234.28	225.90	216.58	216.42	222.87	220.38	209.74	242.31	215.73	212.03
F1	M	337.26	323.97	361.12	402.76	470.07	796.11	677.76	359.45	409.11	507.31
	W	370.19	330.16	420.28	456.73	538.88	685.73	723.2	405.28	4748.33	652.91
F2	M	2109.35	2111.51	1832.85	1960.38	1405.86	1349.03	1332.85	1139.35	1049.16	1021.15
	W	2178.39	2295.82	2000.99	2150.34	1514.20	1433.13	1426.14	1203.76	1069.05	1109.55
F3	M	2811.48	2826.04	2564.85	2540.82	2657.60	2747.66	2734.02	2704.18	2934.53	2707.26
	W	3117.37	3258.36	2933.46	2907.49	2993.09	2755.78	2752.37	2945.03	3101.95	2861.29

Figure 1 depicts the distribution of fundamental frequency for the vowels in different preceding consonant contexts (with respect to manner of articulation). Figure 2 shows the scatter plot of the individual values of the different vowel productions along with ellipses fit for each vowel. Figures 3 and 4 depict the acoustic vowel diagrams of vowel space for the mean F1 and F2 for male speakers and female speakers, respectively.

Fig. 1.

Fundamental frequency distribution for vowels in different preceding consonant contexts

Fig. 2.

Scatter plot of different vowel productions along with ellipses fit for each vowel

Fig. 3.

Acoustic vowel diagrams of vowel space for mean F1 and F2 for male speakers

Fig. 4.

Acoustic vowel diagrams of vowel space for mean F1 and F2 for female speakers

Durational Characteristics of Vowels

The vowel duration for each vowel for both the genders has been denoted in the Fig. 5. The long vowels (/i:/, /e /, /ä/, /ɔ/ had longer durations as compared to short vowels. The effect of consonant context (manner of articulation) on the vowel duration has been depicted in Fig. 6 given below.

Fig. 5.

Vowel duration for each vowel for both the genders

Fig. 6.

Effect of consonant context on the vowel duration

Discussion

The information about acoustic characteristics of the speech sounds helps to understand the articulatory nature of its sounds and their perception [3]. Ladefoged [15] states that the traditional articulatory description of the vowel sounds is not satisfactory. He further adds that vowels noted as ‘high vowels’ vary with respect to their height across languages while ‘back vowels’ vary with respect to the degree of their backness. According to him, the high/low and front/back contrasts used for vowels are best described using acoustic terms. In addition, studies have brought to light the differences in the vowel quality using the same transcription and symbols when used across different languages [8, 15]. Thus, it would be interesting to study the spectral and temporal characteristics across different languages and dialects. Spectral and temporal characteristics were studied for vowels in male and female speakers of Mangalorean Catholic Konkani dialect speakers of Konkani. The spectral cues of the fundamental frequency (F0), first three formants (F1, F2, and F3) and temporal cue of vowel duration (in m sec) were explored.

The maximum F0 was observed for the high back vowel /u/ for both the genders (163.63 Hz in males and 242.31 Hz in females). The minimum F0 was observed for front mid vowel /e / in males (143.35 Hz) and central low vowel / ä / in females (209.74 Hz). Variation in F0 was notedbetween both the genders as also found in previous studies in other languages [4, 12, 16]. Higher fundamental frequency values were observed for short vowels as compared to long vowels. Higher F0 was noted for high vowels as compared to low vowels. The back vowels had higher F0 values as compared to the front vowels. These findings are in support of similar findings in other languages such as Telugu [12], English [17] and Hebrew [16].

Further, the data was scrutinized in terms of the effect of consonant context on the F0 values. It was noted that F0 was highest for /i/ vowel in the context of an alveolar-palatal fricative preceding the vowel followed by /u/ vowel with a preceding fricative context. Lowest F0 was noted for / ä/ vowel when preceded by a semivowel and a nasal consonant context. High and mid vowels have higher F0 values as compared to low vowels. The consonant context in which a vowel occurs has an impact on the spectral properties of the vowel sound [18–20]. According to literature [15, 21], the first two formants (F1 and F2) provide important information about the quality of the vowel. These are related to the open/close (high/low) and front back/back aspects of shape and tongue position while producing these sounds.

Thus, the F1 is at a higher frequency for open vowels such as /a/ and lower frequency for close vowels such as /i/. In the present study, among lowest values for first formant are noted at /i:/ for both male and female speakers. While, highest values for first formant are observed for /ɐ/ for male speakers and /ä/ for female speakers. Overall, the open vowels had a higher value for F1 while the close vowels had a lower value. The second formant is at a higher frequency value for front vowels such as /i/ and at a lower frequency value for back vowel such as /u/ [15, 21]. In the present study, the highest second formant values were noted for the vowel /i:/ in both males and females, while least values were noted for the back vowel /ɔ/ in males and /o/ in females. Overall, higher values for F2 were observed for front vowels and lower values for back vowels. Akpanglo-Nartey [22] stated that despite the effect of gender on vowel quality, the difference in males and females are not consistent for all vowel types and thus gender might not be a significant variable for these differences. The vowel diagrams for males and females were obtained by plotting the F2 (Hz) on the x-axis and the F1 (Hz) on the y-axis. These diagrams are suggestive of some overlap present in the formant values for the same vowel when compared between genders. These discrepancies are also noted in the present study, because of which it is difficult to differentiate the vowels between genders solely based on spectral and temporal cues due to the overlapping values.

Central vowels have a higher duration followed by front and back vowels as reported in several languages such as Hebrew [16], Telugu [12], English [23]. Effect of gender has also been noted on vowel durations with females having longer durations as compared to males [6]. Vowels with preceded stop consonants had longer durations followed by nasal and fricative sounds in English [24, 25]. The long vowels (/i:/, /e/, /ä/, /ɔ/) had longer durations as compared to short vowels. The vowel duration for females was longer as compared to males. In females, the longest vowel duration was noted for /e/ and /ä/ of 190 ms while in males, the longest duration of 160 ms was noted for /e/, /ɔ/ and /ä/ vowels. The shortest vowel duration wad noted for females for /ɐ/ and /u/ of 130 ms and for males for /i/ and /ɵ/ of 120 ms. The vowel front mid vowel /e/ has the longest duration in the preceding affricate context. Ladefoged [5] illustrates the differences among vowels in different languages which have subtle differences though might be perceived to be the same.

Conclusion

It is interesting to study these subtle spectral and temporal difference exhibited by different languages across genders and dialects. This acoustic information in terms of spectral and temporal cues would help to better understand the production and perception of languages and dialects. The Mangalorean Catholic Konkani dialect of Konkani spoken in Mangalore, Karnataka, India is one regional and religion based dialect of the Konkani language. This dialect heavily borrows from regional Kannada language as well as the main Konkani language. The findings of the present study serve as a reference point for spectral analysis for speakers of this language. This study would serve as a foundation for further studies on the phonetics, production, perceptual and speech synthesis in this language. Further exploratory studies need to carry out across other consonant contexts, across age ranges, and in clinical populations.

Compliance with Ethical Standards

Conflict of interest

There is no conflict of Interest to declare.