Speech rhythm has remained a controversial topic for a long time, especially in the field of acoustic phonetics. Examples of such controversial issues are (to name a few): a) how to define and measure speech rhythm and whether this concept is valid and/or useful (Arvaniti, 2012; Tilsen, 2016), b) if distinct rhythm classes or types such as stress-timed and syllable-timed languages exist and whether or not they reflect linguistic or perceptual categories (Dauer, 1983; Ramus et al., 1999; White & Mattys, 2007); and c) how speech rhythm relates and interacts with other prosodic or segmental features such as intonation, stress, vowel quality, and syllable structure (Grabe & Low, 2002; Arvaniti et al., 2008; Dellwo et al., 2015). The fact that such issues exist reflects the diverse and complex nature of studies done in this area, which, in turn, have served as the rationale behind the current study.

Moreover, one significant contribution of such studies to the body of literature has been the introduction of the above-mentioned issues in at least four different branches of applied phonetics, namely 1) Children’s speech and second language learning (Lee et al. 2014; Polyanskaya & Ordin, 2015), 2) Speech technology (Barbosaand & Bailly, 1994, Gibbon, 2023), 3) Speech pathology (Leong & Goswami, 2014; Liss et al., 2009; White et al., 2010, Magne et al., 2016) and, 4) Forensic phonetics (Dellwo et al., 2012; Leeman et al., 2014; Asadi et al. 2018).

The distinction between speaker variability and language-dependent rhythmic characteristics is very relevant, as it helps determining the sources and effects of rhythmic variation in speech (Fuchs, 2016). Nonetheless, drawing a clear line between such concepts is not an easy task since they tend to interact and influence each other in complex ways (Mok & Dellwo, 2008). For example, some speaker-specific features may or may not be pronounced depending on the language or dialect spoken by the speaker (Leemann et al., 2014). while, some other language-specific features may be pronounced more strongly or weakly depending on the speaker’s individual style or preference (Asadi et al., 2018).

Consequently, to distinguish between speaker variability and language-dependent rhythmic characteristics, both acoustic measurements and perceptual judgments of speech rhythm are needed. Acoustic measurements provide objective and quantitative data on temporal features of speech like duration, intensity, frequency, and variability (Gibbon, 2023).

That being said, to study the between speaker variability and language-dependent rhythmic characteristics, this study attempted to find out the appropriate measures for between-speaker and between-sentences rhythmic variability in two speaking styles in Kalhori. Kalhori was selected because it has unique features that may affect its speech rhythm, such as its complex syllable structure (C)(C)V(V)(C)(C), its stress pattern (penultimate or final), its vowel harmony system (front-back and round-unround), and its tonal accent system (Karimi-doostan, 2002; Kreynbroek, 2005; Thackston, 2006). Therefore, segmental intervals, consonant and vowel intervals, vocalic and consonantal intervals, voiced and unvoiced intervals, syllable intervals, and syllable peak intervals were examined in both spontaneous and read speech styles of Kalhori, a variety of Kurdish.

Kurdish is a covering term used to refer to a group of Northwestern Iranian languages spoken in parts of Turkey, Iran, Armenia, Iraq, Syria, and Azerbaijan (Windfuhr, 1989). Generally, there is no agreement on the classification of Kurdish dialects whether in Iran or other countries. McCarus (2009), for instance, believes that Kurdish cannot be located in a single group among Iranian languages because, according to Gharib (2011) and McCarus (1959), it shares syntactic and morphological similarities with Balouchi, Gilaki, Taleshi, and Farsi. Dabirmoghadam (2013) Daneshpazhouh (2010), Thackston (2006) and Kreynbroek (2005) provide different classifications for Kurdish. It is, however, mainly divided into three main groups including Northern Kurdish or “Kurmanji”, Central Kurdish or “Sorani” and Southern Kurdish. According to Fattah (2000), Southern Kurdish is spoken by three million people across an extensive region in Kermanshah, Ilam, Parts of Lorestan and Kurdistan Provinces in Iran and Khanaqin and Mandali in Iraq. As Figure 1 illustrates, the southern Kurdish consists of several varieties including Kermashani, Feyli, Laki and Kalhori. The data for the current study has been based upon Kalhori, one of the biggest tribes of Kermanshah and the second biggest tribe in Iran. Kalhori is the spoken variety in Iran’s Kermanshah (in Eslamabad, Gilan-e-Gharb, southern part of Qasr-e-Shirin), Ilam (in Abdanan), Kurdistan (in Bijar and Qorveh) and Iraq (in Khaneqeyn, Kalar, Kofri and Diyala). Figure 1 shows the Revised Map of the distribution of Southern Kurdish dialects (Belelli, 2019).

The current study aims to respond to the following research questions using two different styles: read and spontaneous speech.

Thence, the first question helped with documenting and describing the rhythmic typology of Kalhori, while the second question allowed for the analysis of between-sentence variation by revealing how the rhythmic measures varied across different sentences within same speakers and speech style. Lastly, the third question showed speakers’ consistency and/or flexibility while producing and maintaining their speech rhythm, and the adaptation of their speech rhythm in different sentence structures, contents, and different styles.

Speech rhythm approaches can be roughly divided into three categories: durational, modulation, and prominence (He, 2022). Durational approaches measure the variability of different phonetic intervals in speech, especially vocalic and consonantal intervals (Ramus et al., 1999; Grabe & Low, 2002; White & Mattys, 2007; Dellwo 2009, 2010; Arvantini, 2012; Dellwo et al., 2015). Modulation approaches analyze the temporal envelope of speech and extract the recurring frequencies and phase relationships of different modulation rates such as syllable and stress (O’ Dell & Neimenen, 1999; Barbaso, 2002; Tilsen & Johnson, 2008; Leong et al. 2014; Malisz et al., 2017; Lancia et al., 2019; Gibbon, 2023). Prominence approaches examine the intensity or spectral variability of speech and use them to identify the rhythmic skeleton or pattern of speech (Todd, 1985; Cummins & Port, 1998; Lee & Todd, 2004).

Since this study aimed at exploring the durational aspects of the Kalhori variety rhythm, a brief overview of durational approach is provided. Durational approaches can be traced back to the theories of isochrony in stress-timed and syllable-timed languages. This theory was first proposed by Pike (1945), and James (1938, 1929) who claimed that “stress-timed” languages, such as English, German, and Dutch, had equal/periodic feet and “syllable-timed” languages such as French, Italian, and Spanish, had equal/periodic syllables. Nonetheless, such attempts proved that the isochrony or quasi-isochrony of durational intervals were not observable in several languages (Bertrán, 1999; Dauer, 1983; Pointon, 1980; Roach, 1982). Later on, other measures for speech rhythm were proposed by phoneticians. Standard deviation of vocalic and consonantal intervals (∆C and ∆V) as well as the percentage of vocalic intervals (%V) were examined for each sentence by Ramus et al. (1999) to determine the rhythmic typology of different languages. Data from Ramus et al. (1999) consisted of five 15- to 19-syllable sentences read by four native speakers in eight different languages. Their entire database contained 2,720 syllables, with each language consisting of 340 syllables. The results of this study indicated that English is a stress-timed language while French is a syllable-timed language based on ∆V and %V.

In order to measure durational variability between sequences of vocalic and consonantal intervals, Grabe and Low (2002) introduced the Pairwise variability index (nPVI-V and rPVI-C) in which they examined 16 languages, in each language a native speaker read the original text or translation of the story “North Wind and the Sun”. This story contained 141 syllables in the English version. Assuming that the average number of syllables in each version of each language is about 150 syllables, the total number of syllables examined in this study were 2256 syllables (16×150). Based on the results of this study, English rhythm shows patterns that are more closely aligned with stress-timed languages, while French leans closer to syllable-timed languages.

White and Mattys (2007) studied PVI, ∆C, ∆V, VarcoC, VarcoV and %V in English and Dutch as representative of stress-timed languages and Spanish and French as syllable-times. Their database included 5 speakers from each language that read the text of a short story.

Varco coefficient and the natural logarithm that are other normalization methods on the speech rate were proposed by Dellwo (2009, 2010) by using Bonn Tempo corpus that consisted of 12 German speakers, 7 English speakers and 7 French speakers at the time of the research.

Moreover, Arvantini (2012) investigated the repetition of acoustical information of syllables instead of segmental units by introducing the amplitude envelope measure of rhythm. Arvantini (2012) used 3 different styles in her study: story reading, spontaneous speech and sentence reading. Participants in her research were from six different languages of Greek, English, German, Spanish, Korean and Italian. Eight speakers of each language were present in this research. In the story reading section, the text of the story “North Wind and the Sun”, was recited for about one to two minutes in the form of spontaneous speech style, and in the sentence reading section, 5 sentences were read by each speaker.

Subsequent research has shown that vocalic and consonantal rhythm measures can vary significantly in a language based on the speaker’s performance (Wiget et al., 2010; Yoon, 2010, Loukina et al., 2011; Arvantini, 2012; Leeman et al., 2014). However, Wiget al. (2010) indicated that %V and VarcoV are more variable than nPVI among different English speakers. While Dellwo and Fourcin (2013) proposed that speaker-specific information is also reported in the duration of voiced and unvoiced intervals in the German-Swiss language, Dellwo et al. (2015) suggested that between-speaker variability of speech rhythm measures is robust in different within-speaker situations by considering %V, ∆V (ln), ∆C (ln), ∆peak (ln) based on the speed of speech production organs movement and linguistic structures in which 12 German speakers read seven sentences in five different speech rates: very slow, slow, normal, fast and very fast.

Persian within-speaker and between-speaker differences with different speech rates have been studied by Asadi et al. (2018) where 10 Persian speakers read the story “The North Wind and the Sun” in 5 different speech rates. The results showed that %V is a robust parameter in distinguishing between-speaker factors. Taghva et al. (2021) studied a read text in Persian and indicated that VarcoC and %V are the robust measures in between-sentence differences in which ten Persian speakers read the story of “The North Wind and the Sun”.

Having studied the literature and to the best of the present researchers’ knowledge, no study has yet comprehensively investigated the quantitative rhythmic measures for Kurdish language and its varieties, which, as mentioned in “Section 1”, is being spoken in parts of Turkey, Iran, Armenia, Iraq, Syria, and Azerbaijan (Windfuhr, 1989). To fill this gap, this study examined the between-sentence and between-speaker rhythmic measures in two different speaking styles (read and spontaneous speech) in Kalhori, a variety of Kurdish.

Ten native speakers of Kalhori variety who were originally from the same region (Kermanshah, which is the largest Kurdish-speaking city in Iran [Borjian, 2017]), including five males and five females, participated in this study. Ages ranged between 21 and 40 with a mean of 31.72 years and SD of 8.81. To be of the same social group, all participants were recruited among Shiraz University students.

The experiment took place at Shiraz University’s acoustic room where the researchers were able to use Zoom h4 recorder. The recorder was positioned diagonally around 20cm away from the participants’ mouths using a base.

To move forward with the experiment, two sets of corpora were compiled. Gibbon (2022) had indicated that depending on the styles, the degree of rhythm may vary from being more rhythmical in the rhetoric of public speeches, poetry recitation and reading aloud to being more arhythmical in planning discussions. Therefore, in the first corpus, the participants read an identical story to determine the rhythmic typology of Kalhori variety and to express between-sentence and between-speaker rhythmic variability in Kalhori read speech. Following previous studies (Pellegrino, 2019; Gibbon, 2022; Asadi et al., 2018), in this study, the potential effects of age, style, and speech rate on the rhythmic metrics was eliminated by selecting participants of approximately the same age group (21-40) who read the same story at a normal speed. In the second corpus, participants were interviewed to observe between-speaker rhythmic variability in Kalhori spontaneous speech.

To identify the typology of Kalhori rhythmic features, and determine the between-speaker and between-sentence rhythmic variabilities, we used the measures investigated in previous studies (Ramus et al., 1999; White & Mattis, 2007; Grabe & Low, 2002; Dellwo, 2009, 2010; Dellwo et al., 2012; Dellwo et al., 2015).

Documenting and describing languages, whether they are endangered or widely spoken, has many purposes, from conserving the inherited knowledge of the language community to exploring the range of structures and communication events the human mind can handle (Gibbon, 2022). One aspect of this range is how language relates to other modes of communication, and one feature of this aspect is the specific rhythm patterns of speech that distinguish a language community, along with other regular events in daily life and culture (Gibbon, 2022).

To respond to the first research question (i.e., to study the rhythmic typology of Kalhori rhythm based on the read corpus), ∆C, %V and nPVI-V were analyzed. Ramus et al. (1999) by calculating ∆C, %V showed that English is a stress-timed language and French is a syllable-timed language while stress-timed languages demonstrated a high ∆C by reflecting high C-interval variability and low %V by reflecting high V-interval variability, and syllable-timed languages indicated a low ∆C and high %V. On the other hand, nPVI that were studied by Grabe and Low (2002) classified English as a stress-timed language and French as a syllable-timed language since the variability of consecutive vocalic intervals in stress-timed languages was higher than syllable-timed languages.

Findings of the descriptive analysis of read corpus (Table 3) demonstrate that the Kalhori nPVI-V is 47.36, std of %V is 5.61 and std of ∆C is 0.016. Table (4) compares ∆C, %V of French and English (derived from Ramus et. al, 1999), their nPVI-V (derived from Grabe & Low, 2002) to the finding of this study. These findings are comparable to the outcome of this study since both Ramus et al. (1999) and Grabe and Low (2002) studies used the story of “The North Wind and the Sun” to collect their data.

As lower value of %V shows more variability of vowel intervals, and a lower value of ∆C reflects less variability of consonant intervals (Dellwo, 2010), Table (4) presents that Kalhori Kurdish has less variability of vowel intervals and less variability of consonant intervals than English and French. Moreover, nPVI-V reflects the variability of successive vocalic intervals.

Drawing on Grabe and Low (2002), Kalhori read speech is placed among the stress-timed languages since Table (4) shows that the variability of vowel intervals in Kalhori Kurdish is higher than French, but lower than English. Consequently, the rhythm class of Kalhori Kurdish can be placed between stress-timed and syllable-timed based on the read corpus with the controlled situation in which participants of the same aged group read a story in a normal speed.

Furthermore, conducting the first experiment in read corpus allowed us to investigate the impact of sentence structure on the rhythmic measures of read Kalhori speech. Five measures of rateSyl, ∆SylLn, VarcoC, nPVI-V and %V were selected based on Pearson correlation analysis. The results indicate that only two of these measures (VarcoC and %V) are significantly different between sentences. While VarcoC is a measure of consonantal variability and reflects the degree of variation in the duration of consonantal intervals, %V is a measure of vocalic proportion and shows the percentage of vowel duration in the total duration of the utterance. These two measures are related to the syllable structure and the vowel-consonant ratio of the sentences (Dellwo, 2010). According to the results (Table 6), VarcoC is only marginally significant VarcoC even while showing a low F-value, proposing a small part of the total variation in the data.

On the other hand, %V is highly significant, meaning that the difference between sentences is due to sentence structure rather than random variation. Moreover, %V has a high F-value (5.41), which is indicative of a large part of the total variation in the data. The results suggest that, based on data, %V is the best measure to determine the Kalhori between-sentence variability. In other words, sentences with different structures have different proportions of vowel duration in their total duration. This may be related to the phonological and morphological features of Kalhori, such as vowel harmony, vowel lengthening, and consonant clusters. Hence, the outcome of this study is aligned with the results of Taghva et al. (2021), who showed that VarcoC and %V are robust measures among Persian between-sentence differences.

To respond to the research question probing the most efficient durational rhythmic measures for between-speaker rhythmic variability in Kalhori speech, the read speech style as well as the spontaneous speech style were examined using five rhythmic measures selected by Pearson correlation analysis: ratesyl, ∆Sylln, VarcoC, nPVI-V, and %V. Therefore, a MANOVA (Table 9 and 10) was conducted to examine which rhythmic measure or measures best discriminated between-speakers. The results revealed that:

Therefore, based on this analysis, the rhythmic measures that best discriminated between Kalhori speakers in both read and spontaneous speech styles were %V and rateSyl. These two measures identified individual speakers most effectively based on durational rhythmic analysis. Consequently, the rate of the syllable intervals together with the vocalic proportion of speech are the most useful features for identifying the speakers based on durational rhythmic measures. Findings of this study are in line with the findings of Asadi et al. (2018) and Dellwo et al. (2015) for Persian and German. Asadi et al. (2018) demonstrated the robustness of %V against both sources of within-speaker variability including time-lapsing and speech-rate variability.

In conclusion, the use of durational measures as a forensic cue may have important implications for the situations where speaker identification information is required (Arvaniti, 2012; Leeman et al., 2014; Dellwo et al., 2015; He & Dellwo, 2016; Asadi et al., 2018). Therefore, the findings of this study hold great potential for enhancing speaker identification in diverse forensic cases. Particularly, the identification of %V and rateSyl as the most distinguishing measures between speakers suggests their potential as valuable acoustic-prosodic features for forensic voice comparison tasks.

However, the comparison of the most discriminative measures for between-sentence variability (VarcoC and %V) with those for between-speaker variability (rateSyl, %V) reveals that %V is influenced by both language-specific and speaker-specific factors, which may affect its variability between sentences and speakers. Hence, while rhythmic measures such as %V hold promise as effective discriminators between speakers, their performance can be influenced by factors other than the voice alone, including linguistic peculiarities. Therefore, forensic practitioners must exercise caution in adapting and validating speaker identification models to account for the specific linguistic and contextual characteristics of the language being investigated.

This study thus sheds light on the complex interplay between language-specific factors and speaker identification, highlighting the need for a nuanced and comprehensive approach to ensure the accuracy and reliability of forensic voice analysis techniques. Analyzing other varieties of Kurdish language could also serve as a fruitful area of study for future attempts.