Speaking Fundamental Frequency and Individual Variability in Caucasian and African American School-Age Children

Purpose: The purpose of this study was to examine the speaking fundamental frequency (SFF) and pitch sigma (individual SFF variability) of African American and Caucasian children ages 6 through 8 years. Method: Participants in this study included 63 Caucasian and African American children recruited from 6 urban schools and 1 day care center. All participants passed hearing and speech–language screenings. Spontaneous speech samples for SFF measurement were elicited from each child in a quiet room in the school he or she attended. Results: Results of this study found that there were no significant differences in SFF or pitch sigma as a function of the races or ages studied. It appeared that a single value for each variable could reasonably characterize African American and Caucasian children at ages 6, 7, or 8: 244.8 Hz for SFF (SD = 30.0 Hz) and 2.06 semitones (STs) for pitch sigma (SD = 0.82 ST). Conclusions: This study is the most comprehensive to date on SFF and pitch sigma for African American and Caucasian children ages 6 to 8 years. Results supported previous observations that SFF is stable throughout the prepubescent years. Furthermore, findings also suggest that pitch sigma is stable across the ages of 6 to 8 years, regardless of race.

T he acoustic characteristics of an individual's voice constitute a highly important dynamic of the voice itself.Acoustic characteristics such as fundamental frequency and vocal variability are significant listener cues to personality and identity, whether consciously realized or not (Aronson, 1990;Perkins, 1983).If those acoustic characteristics of a person's voice are outside of the typical range, the voice is considered to be disordered.It is imperative to have normative data available for speech-language pathologists to accurately assess the presence and severity of voice disorders, especially for children in the schools.
Fundamental frequency is an acoustic measure reflecting the rate of vocal fold vibration (Baker, Weinrich, Bevington, Schroth, & Schroeder, 2008).Speaking fundamental frequency (SFF) is the central tendency of the frequency of vibration of the vocal folds during connected speech (Baken & Orlikoff, 2000) and correlates with the perceived pitch of a speaker's voice.Speech, however, is not monotonous.Typical speakers vary their SFF as a function of sentence meaning (reflected in stress patterns), sentence type (e.g., declarative vs. interrogative), and affect.Thus, speechlanguage pathologists are interested in both SFF and the variability of SFF during an individual's connected speech (Baken & Orlikoff, 2000).
Use of an inappropriate SFF can be both a cause and an effect of a voice disorder.An SFF that is too high or too low is considered part of a vocal abuse/misuse syndrome, also known as phonotrauma and can contribute to laryngeal tissue changes and the development of benign lesions (Colton, Casper, & Leonard, 2011).School-age children-in particular, boys ages 0-9 years-have, by far, the highest occurrence of benign vocal fold lesions of any age group (Cohen, Kim, Roy, Asche, & Courey, 2012).The number of children with voice disorders, regardless of etiology, has been estimated as anywhere from 3.9% to 24% of the pediatric population (Carding, Roulstone, & Northstone, 2006;Duff, Proctor, & Yairi, 2004;Powell, Filter, & Williams, 1989).In the large sample of individuals seeking treatment for dysphonia (voice disorders) studied by Cohen et al. (2012), approximately 28% of the boys and 25% of the girls (0-9 years of age) were diagnosed with benign vocal fold lesions.Thus, it can be concluded that about one quarter of the elementary school-age children who present with voice disorders have lesions due to phonotrauma.
Acoustic measurement of fundamental frequency in isolated vowels or SFF in connected speech has become "a standard objective approach in understanding vocal fold vibration due to its noninvasiveness" (Chen, Kimelman, & Micco, 2009, p. 74).Lesions such as nodules and polyps tend to lower SFF values because they increase the mass of the vocal folds.Acoustic measures of fundamental frequency can serve as baseline measures of vocal function before treatment or surgery and can be used to monitor the progress of behavioral intervention (Baker et al., 2008).Comparing the frequency measures of a child with a voice disorder to those of children with typical voices can aid speech-language professionals in determining the contributing and maintaining factors of the voice disorder, as well as in assessing progress as a result of therapy (Chen et al., 2009).Therefore, accurate data on the frequency characteristics of children with typical voices are important in the diagnosis and treatment of children with voice disorders.
Measures of individual variability in SFF are also important in the evaluation and treatment of children with voice disorders.Children with vocal fold lesions may have limited vocal variability due to the lesion's interference with typical vocal fold vibratory patterns through a range of frequencies; in addition, it has been suggested that children with a hearing loss may show inappropriate frequency and intensity ranges (Andrews, 2006, pp. 229-230).Monopitch, or a voice that does not contain the typical amount of perceived variation in pitch expected in connected speech, may be seen in children with neurological impairments or may be a reflection of a psychiatric disturbance (Colton et al., 2011).It has been suggested that children with childhood apraxia of speech may demonstrate "reduced range of or variable (perceived) pitch, as well as reduced range of or variable (perceived) loudness, (which) gives the listener the impression of monotone, monoloud speech, respectively" (American Speech-Hearing-Language Association, 2007, p. 20).Prosodic disturbances have also been noted in the speech of children with high-functioning autism; specifically, greater SFF range in conversational speech compared to typically developing peers (Nadig & Shaw, 2012).Considering all of the disorders that may involve abnormal measured or perceived vocal variability, objective data on frequency variability in typically developing children may help the speech-language pathologist in diagnosing and tracking changes in children with a variety of communication disorders.
One measure of SFF variability is pitch sigma, which is the average distance of all frequency values in an individual's connected speech sample from the mean SFF (Baken & Orlikoff, 2000).Algebraically, this can be expressed as & Orlikoff, 2000, p. 172).
Thus, pitch sigma is the standard deviation of an individual's frequency productions in a given connected speech sample and reflects the degree of an individual's variability in SFF (Awan & Mueller, 1996).It should not be confused with the standard deviation that is calculated when a group of speakers' SFFs are averaged.The latter type of standard deviation represents the degree of variability in the SFFs observed in the group.While pitch sigma is a measure of an individual's variability in frequency production, a group standard deviation is a measure of how much variability there is in the mean SFF among group members.Because "standard deviation" can be used to refer to both measures, individual variability is referred to as pitch sigma.
Pitch sigma can also be differentiated from SFF range, or the difference between the lowest and highest frequencies in a connected speech sample.This is a simple way of describing the degree of SFF variation (Baken & Orlikoff, 2000).However, according to Snell (1995), both analog and digital approaches to measurement can introduce highand low-frequency artifacts in to the analyzed signal.Because of their extreme values, these artifacts are sometimes misidentified as the upper and/or lower limits of the SFF range.The SFF range, therefore, is easily distorted by erroneous upper or lower limit values, since range is defined by the extremes in the sample.However, because all the frequencies produced by the speaker are included in the pitch sigma calculation, pitch sigma is less influenced by extreme values and is not as likely to be affected by artifacts in the sample (Baken & Orlikoff, 2000).
Like all measures of frequency variability, pitch sigma is best expressed in semitones (STs).A semitone corresponds to a half step on the Western chromatic scale (Behrman, 2013).An octave in the Western musical scale is 12 STs; thus, an ST is 1/12th of an octave.According to Behrman (2013, p. 34), the difference in STs between two frequencies equals 39.86 × log 10 (f higher /f lower ).Because hertz values double for each succeeding octave, perception of frequency differences does not always correlate well with hertz value differences.For example, a listener hears the difference between 100 Hz and 200 Hz (a 100-Hz change) as one octave, a clear perceptual phenomenon.However, as we move higher in the frequency spectrum, listeners also hear the difference between 400 and 800 Hz (a 400-Hz change) as one octave.This nonlinearity in the auditory systems creates nonequivalencies when differences in hertz values are used to describe variability in frequency samples in different parts of the frequency spectrum.Thus, STs rather than hertz values are typically used to describe variability in SFF production, so that variability for low versus high SFFs can be compared.STs are sometimes presented in terms of musical notation; for example, C 4 (C in the fourth octave, or middle C).Alternatively, STs are sometimes numbered (ST# 48) according to a system developed by the Acoustical Society of America (1960).To provide maximum interpretability to the reader, this article uses both conventions in reporting SFF and SFF variability data.
There have been several studies over the past decades describing the SFF and variability of prepubescent children in connected speech.Table 1 presents a summary of the existing research on SFF in the connected speech.Only data for ages 5-10 years are included on Table 1, since prepubescent school-age children are of primary interest in this study.In general, these studies found no differences in SFF between boys and girls of ages 5-10 years.The mean SFF found for this age group appeared to be in the range of A 3 to B 3 , or ST# 45 to , for children who were speakers of American English, although some studies showed deviations (e.g., Sorenson, 1989).
Despite the trends seen in the data, methodological issues complicate the generalizability of the studies in Table 1.In addition to different sample lengths among studies, the instrumentation used by the various researchers spanned a wide variety of technologies.Weinberg and Zlatin (1970) used the Fundamental Frequency Indicator, an early hybrid analog/digital device; Sorenson (1989) used a prototype digital system that was a precursor to today's fully digital methods, potentially affecting clinical application.Sorenson (1989) had only 3 participants per gender/age group, making statistical analysis of such groups problematic.Moreover, only Weinberg and Zlatin (1970) included measures of pitch variability in terms of pitch sigma.
In addition to gender and age, the influence of race on SFF and pitch sigma has also been considered in schoolage children.A summary of the existing research on SFF in the connected speech of prepubescent children of different races is presented in Table 2.As in Table 1, only data for ages 5-10 years are included, since prepubescent schoolage children are the focus of this study.Of the three studies presented in Table 2, two made comparisons among children of different races: Awan and Mueller (1996) compared African American, Caucasian, and Hispanic children; and Morris (1997) compared African American and Caucasian children.
With regard to the variable of race, results for Morris (1997) showed no significant differences in SFF based on race (or age).In fact, if the data for his African American and Caucasian participants are averaged over age, the results of Morris (1997) show a mean SFF of 217 Hz (ST# 45 or A 3 ) for both racial groups, similar to the central tendency of the studies listed in Table 1.In contrast, Awan and Mueller (1996) found that Hispanic children had a statistically significantly higher mean SFF (249 Hz, ST# 47, or B 3 when averaged over gender) than African American children (236 Hz, ST# 46, or A# 3 ).Despite the findings of statistical significance, these results again are consistent with the results of the studies cited in Table 1.However, Awan and Mueller (1996) studied only 5-year-olds, leaving many gaps in the data on race.
Pitch sigma as a function of race was studied by both Awan and Mueller (1996) and Morris (1997).Awan and Mueller (1996) found no significant differences in pitch sigma for the races they studied.Morris (1997) found both race and age differences: His study showed significant differences in pitch sigma between 10-year-old African American boys compared to 8-year-old and 9-year-old African American boys, and between 10-year-old African American boys and 10-year-old Caucasian boys.It should also be noted that the pitch sigma values obtained by Awan and Mueller (1996) were almost twice as large as those obtained by Morris (1997).This could have been due to the ages of the children involved (5 years of age for Awan & Mueller; 8-10 years of age for Morris), although Weinberg and Zlatin (1970), who studied 5-and 6-year-old Caucasian children, also found pitch sigma results similar to those of Morris (1997).Thus, the large pitch sigmas found by Awan and Mueller (1996) could relate to an age/development effect, or they could be a function of methodological differences among the studies.
In summary, the frequency and variability values that clinicians should be applying to prepubescent children of different ages and races are not entirely clear.There are gaps in the SFF data for certain ages when comparing African American and Caucasian children.There are no studies of the SFF or pitch sigma of 7-year-old African American children, and the data for SFF and pitch sigma for African American children ages 8 through 10 do not include female participants.
Another important question to emerge from this review involves the development of pitch sigma.The pitch sigma of the youngest children studied (5 years old; Awan & Mueller, 1996) showed an average value of 5.04 ST.Weinberg and Zlatin (1970), who studied 5-and 6-year-old children, obtained an average pitch sigma of 2.95 ST.Morris (1997) found that pitch sigma values ranged from 1.9 ST to 3.2 ST for boys ages 8-11 years.It appears that the pitch sigma values of Awan and Mueller's (1996) 5-year-olds are considerably higher than the values of either Weinberg and Zlatin's 5-to 6-year-olds or Morris's 8-to 11-year-olds.Does this reflect a trend for pitch sigma to decrease with age?If so, does the decrease appear to be gradual as the child ages, or does the change appear to be abrupt?Also, are there any differences related to race, as Morris's results would suggest?
To address some of these issues, this study sought to examine the SFF and individual variability (pitch sigma) of prepubescent African American and Caucasian children of ages 6, 7, and 8 years.Specific research questions were as follows: 1.
What SFF and pitch sigma values characterize Caucasian and African American children of ages 6, 7, and 8 years?Current data suggest that, for SFF, frequency values from 207 Hz to 250 Hz (ST# 44 ) might be expected.Results regarding pitch sigma are more difficult to predict, as fewer studies have addressed this variable, and the results have not been as consistent as those reported for SFF.It is possible that the younger children in this study could have pitch sigmas in the 4.0-5.0ST range (as suggested by the research of Awan & Mueller, 1997), or they could be in the 2.0-3.0ST range (as suggested by the research of Weinberg & Zlatin, 1970).The older children in this study are more likely to have pitch sigmas in the 2.0-2.5 ST range, based on the research of Morris (1997).A trend toward decreasing pitch sigmas with age might be seen, or it is possible that pitch sigma may be stable across the age groups tested.Can a single SFF value and a single pitch sigma value be used clinically for comparison purposes for children ages 6-8 years who are being evaluated for voice disorders?An affirmative answer to this question would require a finding of no significant differences in SFF and pitch sigma based on age or race.Current studies suggest that SFF may not vary as a function of age or race but that pitch sigma might show significant effects for either or both independent variable(s).
Because there have been no reports of significant differences between prepubescent boys and girls in terms of SFF or pitch sigma, gender was not included as a variable in this study.The purpose of studies such as this is to contribute to the establishment of relevant clinical data for speech-language pathologists working with school-age children with voice disorders and to clarify the development of typical SFF variability.

Participants
Participants in this study included 63 Caucasian and African American children divided into three age groups: 6-, 7-, and 8-year-olds.For a child to be included in this study, the child's age at the time of recording had to fall into one of the following groups: 6;0 (years;months) to 6;6, 7;0 to 7;6, or 8;0 to 8;6.We decided to target children only to the half year at each age to maximize age differences between successive groups.Additional criteria for inclusion were as follows: either African American or Caucasian heritage as reported in school records by parents/guardians; native speaker of American English (including African American vernacular English) as reported by parents/guardians; typical speech, language, and hearing abilities as evidenced by passing the appropriate screenings; no evidence of sinus cold, infection, upper respiratory, or allergic symptoms on the day of testing; and no history of cognitive or neurological disorders as reported by parent/guardian.Children who were enrolled in special education programs were not eligible for the study.Special education status was determined through school records.Final participant demographic data are summarized in Table 3.
Participants were recruited from six urban public schools and one day care center in the city of Milwaukee, WI.All contact with participants was approved before the initiation of the study by the Institutional Review Board of the University of Wisconsin-Milwaukee and the Division of Assessment and Accountability of the Milwaukee Public Schools.To begin the recruitment process, a description of the study, an informed consent form, and a brief questionnaire addressing the relevant inclusion criteria were sent home with each child in targeted classrooms (selected based on expected child ages).Potential participants who returned a signed consent form and questionnaire were checked for eligibility according to inclusion criteria and scheduled for a screening and experimental session if all were met.All screening and experimental sessions were scheduled at the school the child attended.

Equipment
The Real-Time Pitch subroutine of the Multi-Speech computer program (Model 3650; Kay Pentax, Montvale, NJ) was used to collect and analyze each participant's SFF and pitch sigma.Participants spoke into a Shure SM58 microphone, which input to an MAudio Buddy dual microphone preamplifier and then into a Dell Latitude D600 (Model PP05L) laptop computer, which ran the Multi-Speech program.The laptop contained a Sigma Tel C-Major Audio sound card.A Radio Shack 33-2055 digital sound level meter was used to assess the ambient noise level of each testing room before each participant's recording.The weight of each participant was taken using a Tanita Ironman Inner Scan body composition monitor (Model BC-554).Height was determined through the use of a tape measure.A Beltone Special Instruments Model 120 audiometer, calibrated within 12 months, was used for the audiologic screening.

Participant Screening Procedures
Before participating in the study, each potential participant was administered an audiologic screening by Sara L. Denor, consisting of an otoscopic check and a pure-tone screening at 20 dB HL at the frequencies 1000 Hz, 2000 Hz, and 4000 Hz (American Speech-Language-Hearing Association, 1997).Passing criteria included no observed occlusion of the external ear canal with cerumen or other abnormalities of the tympanic membrane and/or ear canal, as well as clinically reliable responses to all frequencies at the criterion dB level in both ears.Each potential participant was also given a speech and language screening by one of the investigators based on a protocol adapted from the Virginia School Health Guidelines (2001).During this screening, the children were asked to repeat sentences containing all consonants in English; tell about a favorite TV show, a school experience, or their family or hobbies; and describe likenesses and differences between word pairs (for example, watch and clock; for 7-and 8-year-old children only).Passing criteria included no misarticulations of ageappropriate phonemes; typical voice quality and fluency; successful repetition of all sentences; and, for 7-to 8-yearold children, the ability to define at least one difference between the two words in a pair.Approximately three children failed some part of the screening.These children's participation in the study was terminated, and they were referred to their school speech-language pathologist for further testing.

Experimental Procedure
During the data collection sessions, participants were taken in pairs into a quiet room in the school.Ambient noise level in each room was below 59 dB (re: Weighting Network C, an approximation of sound pressure level).None of the examining rooms was located near high-traffic areas in the school, so that episodic noise from student movement and talking in the halls was minimized.The noise in all rooms could best be described as barely noticeable, lowfrequency, and continuous, relating to the heating/cooling/ ventilation system.Children were weighed and measured and were then asked to sit at a small table in front of the microphone.
After a short training period to familiarize each participant with the experimental task, each participant was asked to tell the investigator about his or her favorite television show or story.This method of eliciting speech samples was similar to that used by Weinberg and Zlatin (1970), Hacki and Heitmüller (1999), and Wheat and Hudson (1988).The objective of the present investigators was to obtain speech samples of adequate length and naturalness so that SFF and pitch sigma results would accurately reflect the child's frequency patterns in nontest speech.Although there are no well-established durations for this, Baken and Orlikoff (2000) have concluded that samples of 14 s are adequate for measurement accuracy of SFF, although durations that are "significantly longer" are necessary to accurately determine SFF variability (pp. 172-173).The present investigators targeted 30-60 s of fluent connected speech, so that 30 s of measurable connected speech could be analyzed for each child.Picture description and other speech elicitation methods that constrain topic were not used, as it was believed they would compromise the naturalness of the child's SFF usage.
Once the children were seated, a head-positioning device made of brightly colored tubes was placed in front of each speaker child in turn, to maintain a constant mouth-tomicrophone distance of 12 in.at a 45°angle.The speaker child was encouraged to speak continuously for 30-60 s.This duration is similar to what an adult speaker would produce when reading the Rainbow Passage (Fairbanks, 1960), a commonly used reading sample in acoustic voice research.While one child was speaking, the other sat at the table and waited for his or her turn.The second child's presence was incorporated into the data-gathering procedure to make the experimental situation more comfortable for both children and to encourage habitual speech patterns, since the speaker was talking not only to the experimenters but also to a peer.
Each sample was digitized at 44 kHz using the Real-Time Pitch subroutine of Multi-Speech and saved with an Impact 2 GB flash drive for later analysis.Two investigators took part in each experimental session.One administered the experimental protocol and sat directly across from the participant to encourage a smooth, continuous sample, and the second stood at the right of the participant to execute the recording process.

Acoustic Analysis
Acoustic analysis for SFF and pitch sigma was performed on each child's entire spontaneous speech sample using the Real-Time Pitch subroutine of Multi-Speech.Average sample length was 36 s (range = 28-50 s).Pitch display range and processing range were set to 125 Hz-625 Hz.To extract SFF and pitch sigma, the Real-Time Pitch program utilized a waveform matching method, whereby the digitized speech samples were divided into 20 ms frames (Snell, 1995) by the program's pitch-processing routines.These frames were analyzed for the presence or absence of voicing and, if voiced, for the frequency of voicing.The pitch smoothing function was then applied to each speech sample to remove extraneous background noise and outlying frequencies.This function eliminated pitch periods that were likely to be artifacts of the recording process.
Following application of the pitch smoothing routine (used to eliminate as many artifacts as possible from the sample), all remaining pitch periods were averaged using the statistics function of Real-Time Pitch.The results for minimum SFF were given particular attention.If the minimum frequency was above 125 Hz, the lower limit of the analysis range, no further processing was done.However, if the speech sample contained frequencies at 125 Hz, the investigators scrolled back through the visual display of the sample to see if the lowest frequencies appeared to be isolated noise bursts (perhaps due to glottal fry) or if they represented the endpoint of a downward pitch glide.If the data point(s) in question appeared to be isolated, the lower limit of the analysis range was reset to 130 Hz, and the analysis was repeated until all lower frequency noise bursts appeared to be eliminated.For each sample, the SFF (in Hz), pitch sigma (in ST), and duration (in seconds) of the sample were recorded.

Statistical Analysis
To address the statistical significance of the effects of race and age on SFF and pitch sigma, and to determine if single values for SFF and pitch sigma could be applied to all of the participants in the study, two univariate analyses of variance (ANOVAs) were completed using the General Linear Models program (PASW; IBM, 2010).Univariate ANOVAs are robust to unequal cells if there are no strong deviations in sample sizes (i.e., one cell twice the size of another; Keppel, 2004).Furthermore, PASW addresses unequal variances, due to unequal sample sizes or other factors, by providing a corrected model to compensate for heterogeneity of var-iance (IBM, 2010); thus, the proposed statistical treatment was considered acceptable.For both ANOVAs, the independent variables were race (two levels) and age (three levels), with a Race × Age interaction included in the design.For the first ANOVA, the dependent variable was SFF (in Hz); for the second, the dependent variable was pitch sigma (in ST).An alpha level of .05 was selected as the criterion for statistical significance.

Results
Results for SFF and pitch sigma for African American and Caucasian children 6, 7, and 8 years old can be seen in Table 4. Visual inspection of the data showed that the mean SFF of African American children, regardless of age, was somewhat lower than the mean SFF of Caucasian children (238.9Hz vs. 250.5Hz).However, when Hz values were converted to STs, this difference had a magnitude of only one semitone (A 3 # or ST #46 vs. B 3 or ST #47).This difference was well within 1 SD of the mean of both groups.Age results for SFF were somewhat more complex.Mean SFF for 6-year-olds, regardless of race, was 240.5 Hz; for 7-yearolds, it was 252.9 Hz; and for 8-year-olds, it was 239.7 Hz.The trend toward a higher SFF for 7-year-olds (compared to 6-and 8-year-olds) was seen in both African American and Caucasian children.However, all differences based on age were within 1 SD of the mean SFF of each age group and, when converted to STs, were only approximately 1 ST in magnitude.
Visual inspection of the results for pitch sigma showed that regardless of age, African American children demonstrated somewhat less pitch variation in connected speech compared to Caucasian children.When children of different ages were grouped regardless of race, 7-year-old children had more pitch variability than either 6-or 8-year-old children.However, similar to the data for SFF, the differences in pitch sigma between the two race and three age groups were well below 1 SD for each group.
To address the statistical significance of the effects of race and age on SFF and pitch sigma, two ANOVAs were completed (IBM, 2010).Results of the first ANOVA, assessing the effects of race and age group on SFF, can be seen in Table 5.There were no significant differences in SFF based on race (p = .072),nor were there any significant differences based on age (p = .197).The Race × Age interaction was also not significant (p = .621).
To better interpret the lack of significance, partial h 2 , a measure of effect size, was also calculated for each independent variable and for the interaction.Partial h 2 is a measure of the proportion of total variance accounted for by each independent variable in a study with the effects of other independent variables and interactions partialed out (Richardson, 2011).It is not sensitive to number of participants (Coe, 2002) and thus provides a slightly different assessment of results compared to traditional statistical significance testing, which is influenced by the number of participants.The partial h 2 value of .056for race suggests that race accounts for 5.6% of the variance observed in SFF.Given more participants, race might become significant but still would only account for 5.6% of the variance in SFF.Age could account for 5.5% (partial h 2 = .055).The interaction between race and age could account for 1.7% (partial h 2 = .017).
Results of the second ANOVA, assessing the effects of race and age on pitch sigma, can be seen in Table 6.Again, there were no significant differences based on race or age or for the Race × Age interaction.The partial h 2 value for race was .045,indicating that the amount of variance in pitch sigma associated with race is 4.5%.The partial h 2 for age was .039,suggesting that the amount of variance in pitch sigma associated with age was 3.9%.Finally, the partial h 2 value for the interaction between race and age indicated that the amount of variance in pitch sigma associated with this interaction was 1.0%.

Discussion
The first research question asked what SFF and pitch sigma values characterize Caucasian and African American children of ages 6, 7, and 8 years and whether there were significant differences in these values based on race or age.Results of this study found that there were no significant differences in SFF or pitch sigma for the races or ages that were studied; thus, a single value for each variable could

Comparisons With Previous Literature
The lack of significance for age and race was not surprising, given previous literature.Morris (1997) and Sorenson (1989) also found no significant differences in SFF based on age, and Bennett (1983) found a significant difference in SFF between only two of her groups (8-and 9-year-olds but not 9-and 10-year-olds).Although it is generally agreed that SFF decreases as children mature and as the vocal mechanism becomes larger (Boone, McFarlane, & Von Berg, 2005), it may be that a wider age range might be needed to see significant changes in vocal pitch as a function of age.This speculation is supported by the findings of Hacki and Heitmüller (1999), who studied children ages 4 to 12 and did find a significant main effect for age.Similarly, previous researchers have not found significant SFF differences between African American and Caucasian children (Awan & Mueller, 1996;Morris, 1997), although Awan and Mueller (1996) found significant differences between African American and Hispanic children's SFFs.
The results of the present study are also in good agreement with the values obtained in several previous investigations of SFF in school-age children.The present study's finding of an SFF of 244.8 Hz (SD = 30 Hz) for 6-through 8-year-old African American and Caucasian children was similar to Weinberg and Zlatin's (1970) findings of approximately 247 Hz for 6-year-old Caucasian girls and boys, with both studies' results converting to B 3 or ST #47.The present results were also similar to Bennett's (1983) results for 8-year-old Caucasian boys and girls, at 234 Hz and 235 Hz, respectively (both convert to A# 3 or ST #46).Morris (1997) found an SFF of 230 Hz (A# 3 or ST #46) for 8-year-old African American boys and an SFF of 213 Hz (G# 3 or ST #44) for Caucasian boys of the same age, the latter being somewhat low compared to the present study, although the present study's results for African American 8-year-olds were similar.Finally, Wheat and Hudson (1988) showed SFFs of 219.5 Hz for African American boys and 211.3 Hz for African American girls.Their results for African American boys are in concert with those of the present study (A 3 or ST #45, compared to the present study's finding of B 3 or ST #47), although their results for girls are somewhat lower than the present study's results (G# 3 or ST #44).
The results of the present study did not agree as well with the specific SFF findings of either Hacki and Heitmüller (1999) or Sorenson (1989), both of which found considerably higher SFFs characterizing Caucasian children.However, Sorenson (1989) had only 3 participants per age/gender group, and Hacki and Heitmüller (1999) studied German children; perhaps these differences compared to the current study could account for a portion of the discrepancy in results.
Comparisons to previous research on pitch sigma are limited by the small number of earlier studies investigating individual pitch variability in children.Awan and Mueller (1996) included pitch sigma in their data but studied only 5-year-old children, and their finding of a pitch sigma of approximately 5 ST stands as the greatest individual variability measured in any of the cited studies.Morris (1997) also included pitch sigma in his design, but the only overlap between his participants and the participants of the present study was for 8-year-old African American and Caucasian boys.Morris (1997) found a pitch sigma for these groups of 2.1 and 2.5 ST, respectively.This compared well with the present finding of 1.96 ST.Weinberg and Zlatin (1970) had overlap with the present study for 6-year-old children.Weinberg and Zlatin's (1970) participants showed approximately 1 ST greater variability (3 ST) compared to the present study's 6-year-olds (1.96 ST).
The results of the present study, combined with previous studies, suggest that pitch sigma in children 5-10 years old is approximately 2-3 ST.The hypothesized age trendspecifically, a reduction in pitch sigma from the ages of 5 years to 10 years, based on the results of Awan and Mueller (1996) and Morris (1997)-was not supported.The results of partial h 2 for pitch sigma in this study indicate that, even in studies with greater numbers of participants and a significant effect for age, the amount of the variance in pitch sigma accounted for by age would be modest (3.5%-4.1%).Awan and Mueller's (1996) finding of a pitch sigma of 5.0 ST for 5-year-old children stands apart from those of other investigations of this area and cannot be accounted for by age trends.However, it is notable that, of the studies identifying 2-3 ST as the characteristic pitch sigma of the subjects, all of them (in addition to the present study: Morris, 1997;Weinberg & Zlatin, 1970) used speech samples of 30 s or greater, whereas Awan and Mueller (1996) used 14-s samples.It may be that, to determine individual frequency variability in connected speech, a sample of 30 s or greater is necessary.

Limitations of the Present Study
There are several limitations to the generalizability of this study.First, the investigators were not able to ensure equal numbers of participants in all race/age groups, and a larger number of girls than boys participated.Although we did not expect to see SFF differences between boys and girls, subtle differences in performance between the genders (e.g., tendency to talk in a natural manner to a female examiner) may have affected our results.Furthermore, the number of participants was somewhat low for a study of this nature.These limitations were due to the difficulty in obtaining signed consent forms at multiple schools in an urban school district, scheduling students at times that would not interfere with academic activities during the school day, and finding children who met the various criteria for the study.Future studies might use multiple school-based research teams to increase opportunities to enroll appropriate participants.
In addition, given the young ages of the participants in this study, the engagement of the children varied in speaking about their topic (favorite television show or book).This study was designed to elicit natural-sounding 30-to 60-s samples of connected speech that contained a realistic representation of the child's frequency use, but each child's approach to the task differed.Many children were excited and animated when providing their speech sample, whereas some children were possibly shy or nervous when speaking about the topic.These differences in engagement could have affected the data that were collected or (less likely) could have masked possible differences in SFF or pitch variability based on race or age.However, a mix of animated children and reticent children will undoubtedly be seen in the typical school setting, so it could also be argued that these data reflect an appropriate cross-section of speech behaviors.
Finally, the race of the investigators may have affected the results of the study.Both investigators were Caucasian, and although neither one observed systematic differences in the speaking behavior of the African American children compared to the Caucasian children, subtle effects based on examiner race may have been present.To try to counteract this problem, the investigators brought children to the examining room in pairs to help create a more natural speaking situation.Still, the effects of examiner race may have affected the results of this study.
Despite the aforementioned limitations, this study provides important data on SFF and pitch variability in young school-age African American and Caucasian children's spontaneous speech.The experimental task and setting were similar to what would be typical for a school-based speechlanguage pathologist in an urban setting when evaluating children for the presence of communication disorders.The equipment used is commercially available and can be found in many clinics, school systems, and hospitals.Future research involving children of other races; larger numbers of children; and, perhaps, other acoustic measures of voice would help enlarge the reference against which potential clinical samples could be compared.

Conclusion
As computer-based acoustic analysis becomes more common in voice evaluation (Chen et al., 2009), it is increasingly important to have data available that can be used by speech-language pathologists for comparison between an individual being evaluated and what is characteristic of others similar to him or her who are not suspected of having a communication disorder.Measures of SFF and its variability can be used to support clinical judgments across a wide variety of ages and disorders.Although no single study can address the needs of the field, lines of converging evidence can guide clinicians in deciding which measures are normal and which are abnormal.The present study contributes to the growing body of knowledge of what is typical in the prepubescent voice.

Table 1 .
Previous research in speaking fundamental frequency (SFF ) and individual variability in elementary school-age children (5-10 years).
Note.Participants' race in these studies was not specified.Semitone (ST) values are given in both musical notation and as numbers established by the Acoustical Society of America(1960).Data from previous research are presented as they were in the original studies for boys and girls at different ages.When there were no significant differences found, we averaged each study's group data for comparison purposes.Gelfer & Denor: Speaking Fundamental Frequency and Variability 399 Downloaded From: http://ajslp.pubs.asha.org/pdfaccess.ashx?url=/data/journals/ajslp/930639/ on 06/16/2017 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

Table 2 .
Wheat and Hudson (1988)F and individual variability in elementary school-age children (5-10 years) as a function of race.Note.Data from previous research are presented as they were in the original studies for boys and girls of different races and ages.BecauseWheat and Hudson (1988)found no significant differences, we averaged the data for the two gender groups for comparison purposes.ST values are given in both musical notation and as numbers established by the Acoustical Society of America(1960).AA = African American; C = Caucasian; H = Hispanic.

Table 3 .
Characteristics of participants.

Table 4 .
Results of SFF measures and pitch sigma based on race and age.Note.SFF SDs are given in hertz values and represent intersubject variability in mean SFF.Pitch sigma values represent individual, or intrasubject, variability (or how much variation in semitones an individual speaker might typically produce in a given speech sample).Pitch sigma SDs represent intersubject variability in pitch sigma.Pitch sigma results are presented in semitones.

Table 6 .
Analysis of variance assessing the effects of race and age on pitch sigma (in semitones) or individual pitch variation.

Table 5 .
Analysis of variance assessing the effects of race and age on SFF (in hertz).