Formant transitions in varied utterance positions

Abstract

Aim

Acoustic characteristics associated with varied utterance positions were examined to understand the acoustic consequences of potential articulatory changes near utterance boundaries.

Methods

Second formant (F2) transition characteristics, including transition duration (ms), transition extent (Hz), and derived slope of transition (Hz/ms), of twelve healthy speakers of American English were examined for two diphthong transitions in sew and sigh and one consonant-vowel (CV) transition in bee in utterance initial, utterance final, and utterance end positions. Speakers performed a task of contrastive stress variation that served to demonstrate the changeability of acoustic characteristics as an index of articulatory change in shaping the vocal tract.

Results

Contrastive stress, as compared to words spoken without increased stress, was associated with longer transition duration, greater transition extent, and a decreased slope. Although some utterance position effects were present, no systematic differences consistent with boundary strengthening or declination could be concluded.

Conclusion

Findings suggest that varied utterance positions may be associated with stimuli-dependent variation in articulatory changes that is reflected in the acoustic output. These results indicate the need to further understand the construct of utterance-level speech materials, such as carrier phrases, in clinical practice and research.

Introduction

It has been demonstrated that articulatory gestures may change with various naturally-existing changes at the utterance level, including prosodic variation. Articulatory changes associated with utterance-level prosodic variation has been described in the experimental phonetics literature in terms of the extent of phonological feature enhancement [1, 2], coarticulatory resistance [3, 4], and the magnitude of the displacement of articulators [1, 2, 4–9]. The linguistic unit of interest for such changes is often on the prosodic domain, defined as “the organizational framework that measures off chunks of speech into countable constituents of various sizes” [10]. Increased articulatory change, or “prosodic strengthening,” has been reported for the boundaries of prosodic domains—the onsets and offsets of prosodic units (such as breath groups) [2, 4, 8]. In this manner, prosodic strengthening has been suggested to facilitate information processing by aiding in marking linguistic units [1].

Articulatory change also has been reported for the end of utterances, which has been associated with the phenomena of phrase-final lengthening and articulatory declination [11, 12]. Phrase-final lengthening is a well-documented phenomenon that describes an increase in segment duration at the end of an utterance as compared to the same segment’s duration in other utterance positions [11,13]. On the other hand, articulatory declination, analogous to the well-documented fundamental frequency (F0) declination throughout an utterance [14–16], suggests that the strength or precision of articulation progressively decreases throughout an utterance [12, 17, 18]. In sum, the end of an utterance may be associated with articulatory characteristics that are different from other parts of the same utterance.

The concept of articulatory change has been applied clinically in a number of ways. For example, strategies of contrastive stress (e.g., contrastive stress drill) and clear speech have been implemented as techniques to improve speech intelligibility [19]. These techniques have been reported to induce more exaggerated articulatory gestures, as inferred from acoustic measures, when compared to typical (or habitual) speech [20, 21]. Another example is the Lee Silverman Voice Treatment^® [22, 23], which in principle induces greater articulatory gestures as a byproduct of voluntarily increased vocal intensity. Therefore, strategies aiming to elicit greater articulatory change in clinical application often require speakers to make voluntary changes to their speech style.

Evidence from the experimental phonetics literature, as reviewed above, would seem to suggest that varied utterance positions may have the potential to naturally elicit change of articulatory gestures. It is expected that changes in individual articulatory gestures are reflected in systematic changes in vocal tract configuration that affect speech output [24]. However, the literature on prosodic strengthening and relevant issues has focused mostly on the articulatory-acoustic representation [25], and the acoustic measures explored have been limited to, for example, F1 and F2 “slice-in-time” measures conventionally associated with phonological features such as tongue height and advancement [1, 4, 25, 26]. Thus, an account of the acoustic consequences of articulatory change associated with varied utterance position is warranted.

To this end, the present study examined acoustic characteristics associated with positions toward the beginning and the end of an utterance. Two diphthongs and their second formant (F2) transitions were chosen as the stimuli for examination because they are associated with more time-varying and extensive articulatory gestures [21, 27–29] when compared to monophthongs. Furthermore, formant transitions reflect the outcome of multiple articulatory gestures combined in shaping the vocal tract as a function of time [24, 30]. Measures of transition characteristics examined were transition duration, transition extent, and slope. The utterance positions of interest included one utterance initial boundary position and two utterance final positions. Utterance final positions were comprised of one position near the end of an utterance (i.e., utterance final) and one boundary position at the very end of the utterance (i.e., utterance end). The interest in exploring a final position and a boundary ending position was motivated by the potential interaction of prosodic strengthening and declination. Strengthening at the right edge of the prosodic unit (the utterance end position) might be expected to offset articulatory declination from the left to right boundary of the utterance.

A contrastive stress condition was included as verification of the sensitivity of the acoustic measures to the overall articulatory changeability in shaping the vocal tract. Contrastive stress has been demonstrated to elicit changes in formant transition extent and possibly transition duration [20, 21].

Three research questions were addressed. First, was contrastive stress associated with longer transition duration, greater transition extent, and increased slope, regardless of utterance position? Second, because the utterance initial position is presumably associated with boundary strengthening and minimal effects of articulatory declination, would it be associated with longer transition duration, greater transition extent, and increased slope as compared to both the utterance final and utterance end positions? Finally, were the transition characteristics for the utterance final and utterance end positions comparable or distinct from one another?

Methods

Participants

Six female and six male native speakers (n = 12) of American English born and raised in southern Wisconsin (below a hypothetical line running through Steven’s Point, WI, roughly paralleling the southern border of the state), within the age range of 20 to 30 years (M = 24, SD = 2.9), and without self-reported history of any speech, language, or hearing disorder, participated in this study. The geographic restriction was employed in an effort to control for dialectal variation, which can have a profound effect on diphthong production [31]. Participants were recruited through e-mail listings and flyers posted on the University of Wisconsin-Madison campus. The experiment protocol was approved by the Institutional Review Board (IRB) at UW-Madison. Each participant signed an informed consent and was naïve to the purposes of the study. Participants were compensated for their participation.

Speech Materials

Stimuli included two diphthong transitions in target words sew and sigh as well as one consonant-vowel (CV) transition in bee. The CV transition in bee was included for comparison with the diphthong transitions. CV transitions have relatively small variation in transition duration [32], and perhaps transition extent (for a given following vowel), and this allows them to serve as a potential control stimulus for comparison to the diphthongs, whose transition durations and extents are known to be quite variable [21, 27–29, 33]. Several factors contributed to the stimuli selection, including controlled initial consonant context for both diphthongs of primary interest, consonant-vocalic combinations associated with real words that may be used in reasonably meaningful sentences, and ease of measurement. It should be noted that the experimental stimuli initially included a third diphthong transition in target word row, but data for this transition was excluded from further consideration after initial analysis due to extremely high variability. Target words were embedded in a total of nine grammatical sentences of medium length (Table 1) to allow potential position-related effects on articulatory gestures to unfold to a fuller extent as compared to shorter sentences. Three utterance positions were studied, including the utterance initial, utterance final, and utterance end positions (Table 1). The utterance initial and final positions were near the beginning and end of an utterance (sentences 1–3 and 4–6, respectively, in Table 1). To hold the immediate phonetic context constant, the target words in the utterance-initial and utterance-final positions were surrounded by schwas. The third utterance position was the utterance end position (sentence 7–9 in Table 1), which was the utterance ending boundary position. To control for the immediate phonetic context, the target word followed a schwa in this position. Participants were instructed to produce the article (i.e., “a”) preceding and/or following the target words in a casual way, to elicit schwa instead of a more formal /ei/. All participants followed the instructions and only used schwa in all experimental recordings used for data analysis.

Table 1.

Speech materials

Utterance Position		Word Count	Syllable Count
Initial	1. A sew above the fine edge covered the hole.	9	11
	2. A sigh about the new plan showed his concerns.	9	11
	3. A bee above the red oak annoyed the bear.	9	11
Final	4. Kate was working to repair it with a sew above.	10	13
	5. Sean was talking to avoid hearing a sigh again.	9	13
	6. The young kids were careful to notice a bee above.	10	13
End	7. Frank was funny to say he needed a sew.	9	11
	8. Joe was silly to think I produced a sigh.	9	11
	9. He was willing to say he wanted a bee.	9	11

Procedure

Participants performed a sentence reading task while seated inside a sound-treated booth. Participants were instructed to read the sentences naturally, at a comfortable rate and vocal volume. Ten repetitions of each of the nine sentences were obtained, with the order of sentences randomized. Recordings were made directly onto a computer with a sampling rate of 22.05 kHz.

A contrastive stress condition was included solely for the purpose of demonstrating that formant transitions were “adjustable” in principle and that the acoustic measures for the stimuli would change in the expected manner as reported in previous studies [20, 21]. The contrastive stress condition was obtained only for selected stimuli and for a reduced number of repetitions (five per sentence) as compared to utterances probing the position effect. The contrastive stress variation was obtained after the recordings of the ten repetitions of the experimental stimuli and only for the utterance initial and utterance end positions. Participants were asked to produce each sentence as if they were clarifying specific information within a given sentence. For example, a participant said “A sigh (stressed) about the new plan showed his concerns” when prompted by “What about the new plan showed his concerns?” All participants successfully followed these instructions.

Measurement

Three characteristics of the diphthong and CV F2 transitions were measured. As stated earlier, the acoustic output specifies the change of articulatory gestures at the level of the vocal tract shape, rather than for a single articulator. Acoustic measures were made with a multiple-purpose speech analysis program, TF32 [34]. Both wideband spectrograms and linear predictive coding (LPC) analysis were used to generate formant tracks [35]. Analysis bandwidth (default = 300 Hz) for each speaker was modified as needed to optimize the resolution for formant tracks. The complete diphthong/vowel nucleus was defined as the period of time from the first discernible (that is, clearly definable) glottal pulse to the last discernible glottal pulse [35, 36]. The first three formant trajectories of the entire diphthong/vowel nucleus were measured and manually corrected, when needed.

The F2 transition of interest was further defined for the major diphthong or CV transition. The onset and offset of the major transitions were marked from the estimated lowest/highest to the estimated highest/lowest F2 points, as shown in Figure 1. Measures included transition duration, transition extent (the difference between the onset and offset frequencies), and the derived slope of transition (transition extent divided by duration) [33]. Transition extent is a measure of the extent of change in vocal tract configuration (and thus of the magnitude of the change), and transition slope is a measure of the speed of this change [24].

Figure 1.

Intra- and inter-experimenter reliability (the latter employing a trained measurer) for transition duration and extent were estimated by comparing replicate measurements of a set of a randomly selected 144 tokens, or 10%, of the total number of measurements from the typical stress condition. The mean absolute difference between measurements for transition duration was 6.6 ms for intra-experimenter error and 6.41 ms for inter-experimenter error. For transition extent, the mean absolute difference between measurements was 1.45 Hz for intra-rater and 0.9 Hz for inter-rater. Intra-rater and inter-rater Pearson correlations for both measures were all greater than .96. The measurement errors fell well within those previously reported in the literature for similar measures [37, 38].

Statistical Analysis

Planned comparisons in the form of paired two-sample t tests with trimmed means (i.e., excluding the smallest and greatest values, to control for outliers) of all twelve participants were performed to examine the differences between conditions and among positions [39–41]. This analysis method was chosen given the research agenda and the absence of a theoretical interest in any interaction effects. For each target word, two contrastive stress comparisons and three utterance position comparisons were of interest. This yielded a total of five comparisons per target stimuli: (1) utterance end vs. utterance end with contrastive stress, (2) utterance initial vs. utterance initial with contrastive stress, (3) utterance final vs. utterance initial, (4) utterance end vs. utterance initial, and (5) utterance end vs. utterance final. Thus, for measures of transition duration, transition extent, and derived slope, respectively, there were a total of fifteen pairs of comparisons. Type I error was controlled by treating each acoustic measure as a family of analysis. An alpha level of .05 was assigned to each family of comparisons, yielding an alpha level of .05/15 = .003 per comparison.

Results

Transition Characteristics: Contrastive Versus Typical Stress

The first research question addressed the sensitivity of the acoustic measures to articulatory changeability, under conditions expected to produce articulatory change. Descriptive statistics in Table 2 indicate a tendency for increased transition duration, greater transition extent, and decreased slope associated with contrastively stressed vocalic segments. For /Оʊ/ in sew, only one comparison was statistically significant, with the utterance initial position with contrastive stress yielding longer transition duration as compared to with typical stress (t (11) = −4.71, p < .001). Nonetheless, descriptively, ten of twelve (10/12) speakers also produced greater transition extent with contrastive stress in the utterance end position. For /ɑɪ/ in sigh, there was an increase in transition duration for /ɑɪ/ with contrastive stress in the utterance end position (t (11) = −4.5, p < .001) as well as the utterance initial position (t (11) = −5.11, p < .001) (Table 2). While the increase in transition extent was not significant, 10/12 speakers produced /ɑɪ/ with greater transition extent with contrastive stress in the utterance initial position, and 11/12 speakers produced greater transition extent with contrastive stress in the utterance end position. There was significantly decreased slope in the initial position as compared to the typical stress production (t (11) = 3.92, p < .003). No contrastive stress comparisons were significant for the CV transition in bee even though descriptively 10/12 speakers produced greater transition extent with contrastive stress in the utterance initial position.

Table 2.

Summary of descriptive statistics

Target Word	Utterance Position	Mean Transition Duration (ms)	Mean Transition Extent (Hz)	Mean Slope (kHz/ms)
sew	Initial	134.50 (28.13)	−501.44 (191.69)	−2.92 (5.30)
	Initial (stress)	193.06 (50.29)	−573.49 (129.01)	−3.18 (0.91)
	Final	133.51 (21.65)	−561.45 (85.29)	−4.34 (0.88)
	End	172.25 (37.83)	−504.80 (147.25)	−2.96 (0.75)
	End (stress)	213.68 (36.03)	−576.20 (149.57)	−2.73 (0.64)
sigh	Initial	160.02 (36.29)	607.46 (114.31)	3.94 (0.79)
	Initial (stress)	207.33 (39.78)	674.63 (125.14)	3.35 (0.59)
	Final	209.84 (34.44)	756.03 (97.13)	3.72 (0.69)
	End	177.35 (51.36)	560.31 (232.04)	3.10 (1.04)
	End (stress)	228.63 (44.41)	690.14 (210.44)	3.03 (0.78)
bee	Initial	91.97 (26.88)	309.47 (170.20)	4.00 (1.86)
	Initial (stress)	111.88 (43.29)	399.40 (162.80)	3.90 (2.10)
	Final	75.54 (15.07)	246.81 (140.54)	3.44 (2.07)
	End	120.01 (28.38)	363.17 (131.61)	3.12 1.31)
	End (stress)	145.11 (41.40)	374.09 (186.79)	2.96 (1.92)

Note. Standard deviations are shown in parentheses.

Transitions Characteristics with Variation in Position-in-Utterance

The second research question addressed the potential effects of utterance initial strengthening as reflected in the acoustic output, and descriptive statistics in Table 2 show that transition characteristics varied by position and by stimuli, but the effects were not consistent. It is noteworthy that effects of phrase final lengthening were present. For transitions in sew and bee, increased transition duration was noted for the utterance end position relative to both the utterance initial and utterance final positions (Table 2). For sigh, increased duration occurred for both the utterance final and utterance end positions relative to the initial position (Table 2). Only two comparisons, both for /ɑɪ/ in sigh, were statistically significant. /ɑɪ/ in the utterance final position had longer transition duration than in the initial position (t (11) = 3.98, p = .002) and had greater transition extent as compared to the initial position (t (11) = 3.85, p < .003).

The third research question explored the possible effects of strengthening and declination interacting toward the end of an utterance, and descriptive statistics in Table 2 suggest different transition characteristics for the utterance final and utterance end positions. For /Оʊ/ in sew, the utterance final as compared to the utterance end position had shorter transition duration, greater transition extent, and an increase in slope, but only the increase in slope was statistically significant (t (11) = 11.11, p < .001). For /ɑɪ/ in sigh, the utterance final as compared to utterance end position had longer transition duration, greater transition extent, and a decrease in slope. No comparisons were significant. As for the CV transition in bee, the utterance final position had significantly shorter transition duration (t (11) = 4.91, p < .001), a significantly smaller transition extent (t (11) = 5.93, p < .001), and an increase in slope (NS).

Discussion

The present study examined acoustic characteristics in varied utterance positions that may reflect articulatory change associated with utterance boundaries. The contrastive stress condition demonstrated the changeability of the articulatory gestures underlying the acoustic measures of the present investigation. The trends revealed in the descriptive statistics and statistically significant comparisons were coherent with the hypothesized increased duration and greater transition extent with contrastive stress, but not with the hypothesized increase in slope. Utterance position effects were present, although the effects were not consistent and varied by utterance position and stimulus.

Particularly, the utterance final and end position effects are not straightforward to interpret. For /ɑɪ/ in sigh, the utterance final as compared to the utterance end position was associated with longer duration, greater extent, and decreased slope. Transition characteristics for /Оʊ/ in sew in the utterance final as compared to the utterance end position included shorter duration, greater extent, and greater slope. For the CV transition in bee, the utterance final versus end position was associated with shorter duration, smaller extent, and increased slope, the opposites of those for /ɑɪ/. These patterns suggest that boundary strengthening and declination effects may interact and cannot be easily teased apart.

A reasonable preliminary conclusion from these findings would be that utterance positions do elicit changes in the acoustic output associated with articulatory change without voluntary changes made by the speaker, although these effects may be stimuli-dependent. This has potential importance for the use of speech materials in research and clinical practice given that utterance position may be associated with natural variation in speech characteristics. More specifically, the design of the widely used carrier phrases ought to be given more thought.

Carrier phrases are widely implemented as part of the speech materials in various aspects of speech language pathology often for the purposes of obtaining phonetic inventories [42], controlling for the phonetic context in research [5, 9, 20], and bridging word and sentence level sound productions [42]. The potential utterance position effects would imply that carrier phrases may contribute to utterance-level variation in the speech data and may serve as a modifiable therapy component to facilitate treatment objectives. It is important to note, however, that the utterances examined in the present study were different in length and complexity as compared to conventional carrier phrases. Therefore, the effects of varied utterance positions warrant further study given the wide use of carrier phrases in clinical practice and research.

The present study has some limitations. A small set of phonetic stimuli were examined, and the inclusion of more vocalic as well as consonant stimuli is warranted. The effort to both include only real words and to control for the immediate phonetic context by embedding target words between schwas contributed to the compromised semantic predictability of the utterances, and this may have affected speaker’s production. Further work including both semantically unpredictable and semantically unanimous utterances would be beneficial. As started earlier, the utterances examined in this study were very different from conventional carrier phrases, and a future direction would be to examine conventional carrier phrases as well as longer utterances that hypothetically allow for the potential utterance position effects to unfold.