A comparison of technologies for recording speech and the effects of speaker age

University of Northern Iowa University of Northern Iowa

UNI ScholarWorks UNI ScholarWorks

Honors Program Theses Student Work

2017

A comparison of technologies for recording speech and the A comparison of technologies for recording speech and the

effects of speaker age effects of speaker age

Jacquelyn Marie Knustrom

University of Northern Iowa

Let us know how access to this document bene;ts you

Follow this and additional works at: https://scholarworks.uni.edu/hpt

Part of the Communication Sciences and Disorders Commons

Recommended Citation Recommended Citation

Knustrom, Jacquelyn Marie, "A comparison of technologies for recording speech and the effects of

speaker age" (2017).

Honors Program Theses

. 276.

https://scholarworks.uni.edu/hpt/276

This Open Access Honors Program Thesis is brought to you for free and open access by the Student Work at UNI

ScholarWorks. It has been accepted for inclusion in Honors Program Theses by an authorized administrator of UNI

ScholarWorks. For more information, please contact [email protected].

Offensive Materials Statement: Materials located in UNI ScholarWorks come from a broad range of sources and

time periods. Some of these materials may contain offensive stereotypes, ideas, visuals, or language.

A COMPARISON OF TECHNOLOGIES FOR RECORDING SPEECH AND THE EFFECTS

OF SPEAKER AGE

A Thesis Submitted

in Partial Fulfillment

of the Requirements for the Designation

University Honors

Jacquelyn Marie Knustrom

University of Northern Iowa

May 2017

A COMPARISON OF TECHNOLOGIES FOR RECORDING SPEECH

This Study by: Jacquelyn Knustrom

Entitled: A Comparison of Technologies for Recording Speech and the Effects of Speaker Age

has been approved as meeting the thesis requirement for the Designation University Honors

Date Dr. Lauren Nelson, Honors Thesis Advisor, Communication

Sciences and Disorders

Date Dr. Jessica Moon, Director, University Honors Program

A COMPARISON OF TECHNOLOGIES FOR RECORDING SPEECH

Abstract

The purpose of this study was to compare recordings of speech samples obtained with a

dedicated recording device to those from more readily available devices. Speech recordings are

used by professionals in the field of communication sciences and disorders to identify and

transcribe features of an individual’s speech, regardless of his or her age. This process requires

high-quality recordings, but devices intended to produce such recordings are often expensive, not

easily accessible, and have few uses. This research addressed the following questions: (A) What

combination of microphones and recording devices provides the clearest speech sample based on

a signal-to- noise ratio? (B) Does peak clipping distort recorded speech samples for certain

combinations of microphones and recording devices? (C) Are differences present between the

quality of recordings for different devices dependent on the subject’s age (i.e., child or adult)?

Participants included 4 male and 5 female adults, ages 18-26, and 4 male and 4 female children,

ages 5-10. Participants identified English as their first language, had hearing within normal

limits, and had no communication disorders. Each participant was recorded performing three

speech tasks. Speech was recorded simultaneously on a dedicated recording device, a personal

computer, and an iPad, each paired with an omnidirectional, condenser lavalier microphone.

Recordings were analyzed for signal-to-noise ratio (SNR) and instances of peak clipping. Results

were compared by device and age group. Analyses revealed that the iPad yielded highest average

SNR, but had the most variable values in most comparisons. The dedicated recording device

generally had the lowest average SNR, but was the least variable. Peak clipping was not a

significant factor for any device. No adult participants produced clipped signals, while three

children did. Results suggest that readily available devices such as personal computers and iPads

are appropriate for the types of acoustic analysis performed in this study.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 1

Introduction

This project investigated the quality of recordings obtained with a dedicated recording

device in comparison to more readily available devices. High quality recordings of speech

samples are essential to effective instruction and research in the field of communication

disorders. Professionals in the field of communication disorders need to identify and transcribe

specific details about individuals’ speech. The dedicated recording devices that are intended to

provide high quality signals are not readily available and can be expensive. If recordings from

non-professional devices were sufficient they could be more cost effective and available in more

environments.

Speech samples were recorded on three different devices in combination with three

microphones. A free software application, Praat (http://www.fon.hum.uva.nl/praat/), which

allows analysis, synthesis, and manipulation of speech recordings was used to determine signal-

to-noise ratio and determine the presence or absence of peak clipping in recordings of three types

of speech samples. Signal-to-noise ratio (SNR) is a measure which compares the strength of a

signal to that of the ambient or background noise. Peak clipping occurs when the amplitude of a

sound wave exceeds the range which the equipment can handle. These analyses provided

comparable measures of the quality of the recordings made with different device and microphone

combinations.

Review of Literature

Speech-language pathologists (SLPs) assess and treat many aspects of communication

and swallowing. Included in communication are speech production and fluency, voice,

resonance, language, cognition, and hearing (American Speech-Language-Hearing Association

[ASHA], 2016). In order to optimize their patients’ ability to communicate, SLPs must be able to

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 2

correctly diagnose and efficiently treat communication disorders. Extensive education is needed

to acquire the skills to do this.

Transcription

Phonetics is an area of study that is important for the field of communication disorders.

Phonetics is the scientific study of speech sounds, regardless of languages in which they are used

(Louko & Edwards, 2001; Small, 2016). The sounds of a language, or phonemes, are classified

by their place of production, manner of production, which refers to how the articulators are

configured to achieve the sound, and whether the sound is voiced or unvoiced, referring to vocal

fold vibration (Small, 2016). Speech samples can be phonetically transcribed for analysis using a

specially designed system, the International Phonetic Alphabet (International Phonetic

Association, 1999). Analyses of phonetic transcriptions can help SLPs make a diagnosis of a

speech disorder; however, phonetic transcription of speech presents several difficulties (Louko &

Edwards, 2001).

Clinicians encounter difficulties when trying to transcribe at the same rate as the client’s

speech. In addition, holding the attention of certain client populations and transcribing their

speech simultaneously is challenging. This means that clinicians often record speech samples

and transcribe them after meeting with their clients. When transcribing predetermined passages

or lists, the clinician is aware of the target sounds. This can result in close listening for each

phoneme, which is beneficial, or it can lead to expectation errors in which the clinician perceives

a correct sound that was not produced, which is not beneficial. In contrast, conversational speech

samples can be problematic because intelligibility may be low and the clinician may have limited

knowledge of the target sounds. Correctness is difficult to calculate if the target word is unknown

(Louko & Edwards, 2001). Although transcribing from a recorded sample cannot eliminate these

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 3

challenges entirely, transcribing from a recording has many advantages. Recordings allow

replaying and segmenting shorter sections of the speech sample. When listening to the recording,

transcribing is the only focus of the clinician. Additionally, the clinician can create more ideal

listening conditions such as listening in a quiet setting and listening through headphones. To

achieve the best results and ensure accurate transcription, the clinician needs a high quality

recording (Huckvale, 2009).

Equipment Demands

When selecting equipment for recording speech, one must consider both the microphone

used to obtain the sample and the recording device. The purpose of microphones in voice and

speech research is to convert sound pressure signals to electric signals with characteristics that

are similar to speech (Svec & Grandqvist, 2010). Because this differs from the function of most

microphones, professionals who record speech need to take special considerations.

Omnidirectional microphones have equal sensitivity to sound sources from any direction.

Directional microphones respond differently to different directions. Howard and Murphy (2007)

stated that omnidirectional condenser microphones are best for single source sounds recorded for

research purposes. A critical distance from the source, in this case the speaker, also needs to be

determined. This type of microphone can be connected to a variety of recording devices which

may or may not result in the same quality of recording. Also, voice characteristics differ by

population which may impact the recording (Howard & Murphy, 2007).

The average fundamental frequency for an adult male is 116.5 Hz. For adult females it is

207.7 Hz, while children average 277.2 Hz (Howard & Murphy, 2007). Microphones pick up a

limited range of frequencies. To produce quality recordings this range must be inclusive of all

frequencies that speakers of any age and gender could produce. Internal noise of the microphone

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 4

also should be lower than even the softest possible phonations in anticipation of soft-speaking

clients (Svec & Grandqvist, 2010). Clinicians should also take into account that children may

have trouble sitting still and maintaining the optimal distance from the microphone. Headset

microphones ensure a constant distance and reduce the effect of environmental noise (Barsties &

De Bodt, 2014). In order to accurately compare the quality of recording technologies, optimal

microphones must be paired with the recording devices.

Little research has been done toward determining minimal hardware requirements for

quality speech recordings. However, some varying conclusions have emerged in the current

literature. Vogel and Maruff (2008) analyzed identical speech samples recorded using three

different methods: a dedicated recorder, a disc recorder, and a laptop computer. Each was used

with a different head-mounted microphone. This study showed that specialized hardware was

necessary for analysis of certain acoustic features, whereas the other methods were sufficient for

investigating key features of frequency and timing. Vogel and Maruff concluded that dedicated

recorders were best for recording and analyzing pitch and intensity perturbation. Other methods

such as a laptop computer were sufficient for investigating characteristics such as jitter and

noise-to-harmonic ratio.

Through their investigation of microphone characteristics, Svec and Grandqvist (2010)

were able to compile a list of recommendations for selecting a microphone. They stated that

“different phonation tasks pose different demands on microphones” (p. 365). Inexpensive

microphones may be appropriate in certain situations, but if specifications including optimal

distance, dynamic limits inclusive of all voice levels, and frequency limits suitable for all

speakers are not available, they should not be used for voice and speech research.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 5

Deliyski, Evans, and Shaw (2004) used multiple software systems to analyze acoustic

signals recorded on several combinations of computers and microphones. Their conclusions

supported the National Center for Voice and Speech’s recommendation of professional, head-

mounted, condenser microphones. They asserted that data acquisition systems that cannot accept

this type of microphone should not be used for acoustic assessment and research. This

recommendation was also supported by Barsties and De Bodt (2014), though they indicated a

need for more research to document the recommendation.

Speech-language pathologists perform many analyses that require high quality speech

and voice recordings. These range from phonetic transcriptions of the speech sounds in a sample

to analysis of the acoustic properties of the sample, even to the level of analyzing the properties

of individual speech sounds (Small, 2016). To date SLPs have limited guidance regarding the

selection of microphones and recording devices for different recording purposes. The guidance

that is available often is somewhat dated and does not reflect the most current recording

technologies available with laptop computers or tablets and smart phones. Thus, the purpose of

this research is to compare different types of recording technology and provide SLPs with

information they need to select technology for recording speech samples.

Research Questions

This research will address the following questions: (A) What combination of

microphones and recording devices provides the clearest speech sample based on a signal-to-

noise ratio? (B) Does peak clipping distort recorded speech samples for certain combinations of

microphones and recording devices? (C) Are differences present between the quality of

recordings for different devices dependent on the subject’s age (i.e., child or adult)?

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 6

Hypothesis

The researcher hypothesized that differences in signal-to-noise ratio would not be

statistically significant. With careful selection and pairing of recording technologies and

microphones, little to no clipping was expected. These results would confirm that more readily

available devices are sufficient for obtaining quality speech samples.

Methodology

Participants

Participants in this study included nine adults, five females and four males, and eight

children, four females and four males. One adult participant was excluded randomly from

comparative analysis to keep sample sizes equal. Adult participants’ ages ranged from 18 to 26

and child participants were between the ages of 5 and 10. Institutional Review Board approval

was obtained and its protocols were followed for the use of human participants. Participants were

recruited via email and fliers posted around the University of Northern Iowa’s campus and in the

department of Communication Sciences and Disorders. Recruited adults received a form

describing the study and provided signatures of agreement to participate (see Appendix A).

Parents of child participants signed a permission form and oral consent to participate was

obtained from the children (see Appendix B). Participants were asked to identify their date of

birth, gender, whether or not they have had a hearing screening and the status of their hearing,

their native language, history of speech therapy, and health at time of testing. Parents of child

participants were asked to provide this information. Individuals who had hearing that was not

within normal limits, did not identify English as their first language, or had a communication

disorder were excluded from participation. Present health was addressed to ensure health factors

were not likely to alter the quality of the speech recording. If interested individuals met the

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 7

requirements, succeeding recording sessions involved three speech tasks being recorded by three

distinct microphone and device combinations.

Equipment

The three devices used to record speech samples were the Marantz PMD 660, an HP

2000-2b19WM personal computer with the recording application Audacity, version 2.1.2

http://www.audacityteam.org/, and an iPad generation four with recording app Voice Record

Pro. The Marantz PMD 660 represented a professional, dedicated recording device, whereas the

personal computer and iPad represented more readily available technologies that are capable of

obtaining speech recordings through free applications. After investigating the best microphones

for recording speech, the researcher decided that each device would be paired with an

omnidirectional, condenser lavalier microphone, consistent with Howard and Murphy’s

recommendations. Lavalier microphones feature clips for hands free use. An Audio-technia

AT803 was paired with the Marantz device, a MOVO USB-MI with the personal computer, and

a Shure MVL with the iPad. These microphones were chosen because they fit the determined

criteria, had input compatible with their respective devices, fell within the same price range (i.e.,

$100 to $150), and were rated similarly in quality. The researcher selected lavalier microphones

rather than head mounted microphones because of the need to record to three different recording

devices simultaneously. This was not possible with a head mounted microphone. The titles of the

microphone, app, and recording device combinations will be shortened to Audacity, iPad, and

Marantz in this paper, though these refer to the entire recording system. Each device and

microphone pairing recorded three samples for each participant.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 8

Speaking Tasks

The researcher asked participants to produce three samples: a standard reading passage, a

word list, and a conversational speech sample. Three tasks were decided on to present a variety

of phonation tasks, challenging the microphones in different ways as described by Svec and

Grandqvist (2010). These types of speaking tasks also represented the variety of speech samples

that SLPs need to record when working with clients. The Grandfather Passage, a standard,

phonetically balanced passage used in speech sample acquisition was read by adult participants.

Children were asked to read a more age-appropriate phonetically balanced passage, the

Caterpillar Passage. A list of 94 words was compiled to feature each English phoneme in every

position: initial, medial, and final. The passages and word list are featured in Appendixes C-E.

The variation of speech sounds and their positions in both the passage and the word list ensured

that recording technologies could capture accurately sounds of any frequency and intensity.

Participants were instructed to read the passage at a comfortable rate. When producing the word

list, participants were asked to pause two to three seconds between each word on the list. For

children who were not yet able to read, the investigator read one sentence or phrase of the

Caterpillar Passage at a time and asked the child to repeat the sentence until the child produced

the full passage. Similarly, the investigator read one word from the list at a time and the child

repeated those words to complete the list. For the final speech task, participants engaged in a

conversation with the investigator. They were prompted to describe their favorite vacation. The

target length of the conversational sample was two minutes. If participants did not speak this

long, they were further prompted with questions about their experience, such as “Who did you

go with?” and “What was your favorite activity while you were there?”

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 9

Environment

Of the 18 recording sessions, 16 took place in a therapy room of a university, speech and

hearing clinic. The remaining two were completed in a small, quiet room intended to have

acoustic properties similar to those of the therapy rooms. Participants read or repeated first the

Grandfather or Caterpillar passage, then the word list, and ended with the conversational speech

sample. The speaking tasks were recorded concurrently on the three devices to minimize

participant inconvenience and to provide similar samples for the devices to record. The three

microphones were attached to a lanyard around the participant’s neck at a measured distance of

15 centimeters from mouth to microphone. Participants were instructed to maintain this distance

to the best of their ability and avoid touching the microphones with their hands or the table

surface in front of them.

Analysis

All three devices produced recordings with the .wav file type. At the completion of a

recording session, the recordings from all three devices were transferred to the researcher’s

laptop computer. All recordings were opened in the Praat acoustic analysis software. Praat is a

flexible tool with many different choices for analyzing speech and other acoustic samples

(Boersma & Weenink, 2017; van Lieshout, 2003). For this study, signal-to-noise ratio and

instances of clipping were determined from the provided waveform view and spectrogram.

Signal-to-noise ratio is a measure which compares the strength of a signal to that of the ambient

or background noise. The point of highest signal intensity in each sample was used as the signal

value. The noise value was computed by segmenting the recording into five sections and

averaging an instance of noise from each. The duration of the signal for the passage and

conversational speech sample tasks were divided by five and the first available 0.25 seconds of

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 10

noise from each segment were used for averaging. The 94 words on the list were divided into

five sections and a word was randomly selected from each section. The first available 0.25

seconds of noise following the designated words were used for averaging in each of the

recordings. Noise was considered to be sound produced by the recording device plus ambient or

background noise. Unintentional, non-speech sounds from the speaker or environment were not

included as signal or noise. The most intense signal value was divided by the average of the five

instances of noise to determine signal-to-noise ratio. The signal-to-noise ratios were then

averaged by participant group for each speech task and further analyzed.

Mean and standard deviation of SNR was first computed for the overall group, adults

and children, and by recording equipment for each of the three tasks. The overall means were

compared across the devices within tasks as an indication of overall quality of recording, as

measured by SNR. Adults and children were then separated within tasks and means and standard

deviations were calculated separately for each device. From these values, 95% confidence

intervals were determined. The confidence intervals reflected the standard error around the mean.

These values were plotted to provide side-by-side comparison of ranges for each device and

microphone combination. Following analysis of SNR values, recordings were inspected for

instances of clipping. The researcher originally planned to report SNR and clipping separately

for the adult male and female speakers. The number of participants in each group was small and

further dividing the adult group into males and females was not practical.

Clipping occurs when the amplitude of a sound wave exceeds the range which the

equipment can handle. This may result in a distorted or less clear signal. As a measure to prevent

clipping, sound tests were conducted prior to recording of the three speech tasks. Participants

were asked to count to ten and read aloud the first two sentences of the passage. The volume unit

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 11

meter on each of the devices was monitiored during these tests and adjustments were made if the

signal was entering the yellow or red area of the meter. A display of yellow or red during

recording was not necessarily indicative of an occurrence of clipping. Instances in which the

signal exceeded the analysis window in Praat, resulting in a value display of 1 were considered

clipped (Boersma & Weenink, 2017). The number of instances of clipping was counted for each

recording. Recordings in which an unintentional noise, such as a slamming door, resulted in a

signal of 1 were not counted as clipping. The number of instances of clipping was compared by

device and microphone combination and by sample population.

Reliability

An independent rater reanalyzed 100 of the data points for signal-to-noise ratio selected

at random. The interrater percentage of agreement for this reanalysis was 83%. The Praat

displays intensity values as ± for recorded sounds and maximum intensity of a signal is the

absolute value. The independent rater noted that the examiner recorded the positive value 17

times when the negative value actually reflected maximum intensity. However, this difference

had minimal impact on the findings of the study. Values were calculated to three decimal points

and the difference was greater than .01 in only one instance. The interrater percentage of

agreement using this more lenient criteria was 99%.

Results

Signal-to-Noise Ratio

Passage. The overall means and standard deviations for the recordings of the reading

passage are shown in Table 1. The overall mean of SNR for the recordings captured by the

MOVO microphone and Audacity personal computer app for the reading passage task was

56.176 (SD = 41.810). For the Shure microphone with the Voice Record Pro iPad app the mean

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 12

was 75.117 (SD = 46.327). The recordings from the Audio-technia microphone and Marantz

recorder had a mean of 32.977 (SD = 22.362) for this task. These values are expressed in Table

1. The average signal-to-noise ratio was greatest for the iPad recordings, but its values were also

the most variable. The Marantz produced the lowest average SNR, but was the most consistent

with the smallest standard deviation. These results were consistent when data were divided into

adult and child sample populations.

Table 1: Overall SNR Results for Recordings of the Reading Passages for Children and Adults

Audacity

iPad

Marantz

Mean

56.176

75.117

32.977

Standard Deviation

41.810

46.327

22.362

Table 2 displays the means and standard deviations for SNR computed separately for the

adult and child groups. These findings are also displayed in Figure 1. The adult participants’

recordings from the PC had a mean SNR of 49.784 (SD = 19.771). This resulted in a confidence

interval of 35.51-83.05 and standard error of measurement (SEM) of 14.27. As shown in Table 3,

the recordings of child speakers from the same device had a mean of 61.235 (SD = 56.61),

resulting in a confidence interval of 13.9-108.58 and a SEM of 47.34. The mean for adult

recordings from the iPad was 65.961 (SD = 21.013), resulting in a confidence interval of 55.61-

88.92 and a SEM of 10.35. The child recordings from the iPad had a mean of 84.274 (SD =

62.975), resulting in a confidence interval of 31.62-136.93 and standard error of 84.274. The

Marantz recorder produced a mean of 33.702 (SD = 22.480) for adults. These recordings had a

confidence interval of 55.61-88.92 and a SEM of 16.95. For the children, the mean of Marantz

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 13

recordings in this event was 32.777 (SD = 24.512), resulting in a confidence interval of 12.28-

53.27.

Table 2: SNR for Adults Recorded While Reading the Grandfather Passage

Audacity

iPad

Marantz

Mean

49.783

65.961

33.702

Standard Deviation

19.771

21.013

22.480

95% Confidence

Interval

35.51-83.05

55.61-88.92

16.75-66.21

Standard Error

14.27

10.35

16.95

Table 3: SNR for Children Recorded While Reading the Caterpillar Passage

Audacity

iPad

Marantz

Mean

61.236

84.274

32.777

Standard Deviation

56.616

62.975

24.512

95% Confidence

Interval

13.9-108.58

31.62-136.93

12.28-53.27

Standard Error

47.34

52.65

20.5

Consistent with the overall values, the mean SNR was highest for recordings taken by the

iPad for both adults and children. The standard deviation was lowest for the Marantz in the child

sample, but Audacity with the personal computer produced the lowest standard deviation for

adult recordings. Overall, the child recordings had higher average SNRs across all three

recording apparatuses, and their values were much more variable resulting in higher standard

deviations for each device. The values for adult and children in the passage task are plotted in

Figure 1. As shown in this figure, the confidence intervals for all of the recording options

overlapped for the child participants, meaning there was no clear advantage for any of the

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 14

device/microphone combinations. For the adults, the confidence intervals overlapped for most of

the comparisons. However, recordings made with the iPad did have a better SNR than recordings

made with the Marantz recorder with no overlap of the 95% confidence intervals.

Figure 1: Means and 95% Confidence Intervals for SNR from the Reading Passages

Word List. As shown in Table 4, the mean SNR of all word list recordings taken with

Audacity was 70.582 (SD = 53.213). The iPad recordings of the word list had a mean of 120.979

(SD = 66.194). Word list recordings from the Marantz had a mean of 28.486 (SD = 21.056). The

iPad recordings produced the highest average SNR and were the most variable for this speech

task as well. The Marantz once again had the lowest SNR, but yielded the least variability.

100

120

140

160

Audacity

iPad

Marantz

Children

MarantziPad Audacity

Adults

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 15

Table 4. Overall SNR Results for Recordings of the Word List for Children and Adults

Audacity

iPad

Marantz

Mean

70.582

120.979

28.486

Standard Deviation

53.213

66.194

21.056

As shown in Tables 5 and 6, the researcher also calculated the results for the adult and

child groups separately. This analysis revealed the mean SNR for adult speakers for the word list

recorded by Audacity was 100.307 (SD = 58.813), resulting in a confidence interval of 51.13-

149.48 and a SEM of 49.18. For this task on the same device, the children had a mean SNR of

40.857 (SD = 24.272), resulting in a confidence interval of 20.56-61.15. The iPad word list

recordings of adults had a mean of 166.193 (SD = 62.2). The confidence interval was 114.18-

218.2 and SEM was 52.01. For the children, iPad recordings had a mean SNR of 75.765 (SD =

29.118), resulting in a confidence interval of 51.33-100.02 and a SEM of 75.765. The Marantz

recordings of the word list had a mean of 38.436 (SD = 23.597) for adult speakers, resulting in a

confidence interval of 18.7-58.17 and a standard error of 19.74. The child speaker recordings on

the Marantz had a mean of 18.535 (SD = 12.918), resulting in a confidence interval of 7.73-

29.34 and a SEM of 10.81. The data for SNR analysis of the word list recordings is presented in

Tables 5 and 6 and displayed in Figure 2.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 16

Table 5. SNR for Adults Recorded While Reading the Word List

Audacity

iPad

Marantz

Mean

100.307

166.193

38.436

Standard Deviation

58.813

62.2

23.597

95% Confidence

Interval

51.13-149.48

114.18-218.2

18.7-58.17

Standard Error

49.18

52.01

19.74

Table 6: SNR for Children Recorded during the Word List Task

Audacity

iPad

Marantz

Mean

40.857

75.765

18.535

Standard Deviation

24.272

29.118

12.918

95% Confidence

Interval

20.56-61.15

51.33-100.02

7.73-29.34

Standard Error

20.3

24.44

10.81

When the word list task’s recordings were separated into adults and children both

samples had the highest signal-to-noise ratio from the iPad. Once again, these values were also

the most variable. The Marantz again produced the lowest average SNR for both adults and

children, and was again the least variable. The adult sample had higher mean SNR values on all

three recording systems for this task. The values for adult and children in the word list task are

plotted in Figure 2. As shown in this figure, the confidence intervals for most of the recording

options overlapped for both the adult and child participants. However, recordings made with the

iPad did have a better SNR than recordings made with the Marantz recorder with no overlap of

the 95% confidence intervals. This finding was true for both children and adults for recordings of

the word list task.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 17

Figure 2: Means and 95% Confidence Intervals for SNR from the Word List Task

Conversational Speech Sample. Table 7 shows the overall SNRs for recordings of

conversations with the children and adults. For the conversational speech sample task, the overall

mean SNR for recordings from Audacity was 55.896 (SD = 47.063). The iPad recordings of this

task had a mean of 66.331 (SD = 29.334). The mean for the recordings from the Marantz device

was 28.265 (SD = 32.149). Again, the iPad recordings had the highest average SNR. However,

the recordings taken by the Audacity app for PC were more variable for this event. This set of

iPad recordings was the least variable of the three devices. The Marantz again had the lowest

overall mean of SNR.

100

150

200

250

Audacity

iPad

Marantz

Children

MarantziPad Audacity

Adults

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 18

Table 7. Overall SNR Results for Recordings of Conversations with the Children and Adults

Audacity

iPad

Marantz

Mean

55.896

66.331

28.265

Standard Deviation

47.063

29.334

32.149

When means and standard deviations were calculated separately for adults and children,

the performance of the devices was different. The mean SNR from Audacity for adult speakers

was 66.549 (SD = 59.51). The confidence interval for this set was 11.51-121.59 and the SEM

was 55.04. The Audacity average for child speakers was 46.575 (SD = 34.375), resulting in a

confidence interval of 17.83-75.32 and a SEM of 28.74. The iPad recordings of adult speakers

for the conversational sample had a mean SNR of 62.165 (SD = 29.229), resulting in a

confidence interval of 35.13-89 and a SEM of 27.04. The child iPad recordings had a mean SNR

of 69.976 (SD = 30.923). This confidence interval was 44.12-95.83 and the SEM was 25.86. The

Marantz device produced recordings with a mean SNR of 23.478 (SD = 13.387) for adults,

resulting in a confidence interval of 11.1-35.86 and a SEM of 12.38. For the children, the mean

from the Marantz was 35.315 (SD = 42.881). The confidence interval for children from the

Marantz was 0.54-71.17 and the SEM was 34.77. The data for the conversation recordings is

represented in Tables 8 and 9 and plotted in Figure 3.

The recordings of adult conversational speech samples were the only data set in which

the iPad did not have the highest mean SNR. The mean of the Audacity recordings was highest

for this group, but values were highly variable among adult participants in the sample. The data

for the children was consistent with the trend in the other sets, having the highest mean SNR

from the iPad. However, Marantz recordings were most variable for the children.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 19

Table 8: SNR for Adults Recorded While Participating in Conversations

Audacity

iPad

Marantz

Mean

66.549

62.165

23.478

Standard Deviation

59.509

29.229

13.387

95% Confidence

Interval

11.51-121.59

35.13-89.2

11.1-35.86

Standard Error

55.04

27.04

12.38

Table 9. SNR for Children Recorded While Participating in Conversations

Audacity

iPad

Marantz

Mean

46.575

69.976

35.315

Standard Deviation

34.375

30.923

42.881

95% Confidence

Interval

17.83-75.32

44.12-95.83

0.54-71.17

Standard Error

28.74

25.86

34.77

Figure 3: Adults and Children-Conversation

-40

-20

100

120

140

Audacity

iPad

Marantz

Children

MarantziPad Audacity

Adults

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 20

Also as shown in Figure 3, the confidence intervals for all of the recording options

overlapped for both the adult and child speakers. For the adult speakers the mean SNR was

greatest for recordings made with Audacity on the laptop computer, but the large SEM resulted

in overlap with both the iPad and the Marantz options. For children the mean SNR was greatest

for iPad recordings, but again the SEM resulted in overlap with both the Audacity and Marantz

options.

Clipping

Each recording was examined for the occurrence of clipping and instances were counted.

Patterns were looked for to determine if clipping was more common among adults or children

and males or females. It was also considered that clipping may be more likely to occur in a

particular speech event. Of the 144 analyzed recordings, seven featured instances of clipping, all

from children. Clipping occurred on two recordings of the Caterpillar Passage: the Audacity and

iPad recordings for the same participant. The remaining five instances of clipping occurred in

conversational speech samples: three on iPad recordings, one on an Audacity recording, and one

on a Marantz recording. Child 001 was clipped in four of her nine recordings, child 002 in two,

and child 008 in one. Two of the children were female and one was male. Neither male nor

female adult participants produced signals that resulted in clipping, though the men tended to

produce more intense signals, suggesting greater risk for clipping. The clipping analysis results

for the children are displayed in Table 10.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 21

Table 10. Instances of Clipping across Recordings of the Child Participants

Audacity

iPad

Marantz

Child 001

2: Caterpillar and

Conversation

2: Caterpillar and

Conversation

Child 002

1: Conversation

Child 003

Child 004

Child 005

Child 006

Child 007

Child 008

1: Conversation

Discussion

Recording Technology

One purpose of the present study was to determine what combination of microphones and

recording devices would provide the clearest speech sample based on a signal-to-noise ratio and

absence of peak clipping or signal distortion. Though there were differences in signal-to-noise

ratio for each speaking task across the three recording systems, values varied greatly and in many

cases ranges of values overlapped across devices. There was no one device and microphone

pairing that was clearly superior to the others, suggesting that they are capable of recording

signals of similar quality. The age of the participants and the speaking tasks also were factors a

speech-language pathologist would need to consider. For the reading passage task, the SNR for

the iPad was clearly higher than the SNR for the Marantz device in the recordings from the adult

speakers. For the child speakers, the SNR results across all three recording options overlapped to

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 22

a considerable degree and no one recording option stood out as the best one. One possible

explanation for these differences is that some of the child speakers were not proficient readers

and completed the reading passage task by repeating the sentences rather than by reading the

sentences. This might have resulted in a speech sample with different characteristics than an

actual passage reading. For the word list task the iPad stood out for both the child and adult

speakers as having a higher SNR than the Marantz recorder. As with the reading passage task,

SNRs for the Audacity and iPad recording tended to overlap. None of the devices stood out as

having better SNRs for the recorded conversations. One of the interesting findings was that peak

clipping did not occur in any of the adult speech samples and only in seven instances among the

children.

The findings of this study suggested that readily available devices, when paired with an

appropriate microphone, yielded speech recordings that were as good or better than the

recordings made with the dedicated recording device. Many speech-language pathologists and

researchers in the field have personal computers or iPads, or have those devices available to them

within their work or educational institutions. However, not all of these individuals are likely to

have access to a dedicated recording device. Apps paired with the PC and iPad were free and

easy to use and appeared to provide recordings that were adequate for the phonetic transcription

and some other speech sample analyses. Any recommendations based on the findings of this

study are tentative. One unexpected finding was the high degree of variability across participants

that yielded large confidence intervals. The sources of this variability warrant further

investigation. The Marantz recorder also proved to be more difficult to operate consistently. The

researcher inadvertently recorded 5 of the 48 samples using the MPEG3 format instead of the

.wav format selected for the study. These 5 instances all occurred with the Marantz recorder.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 23

Additionally, newer models of all three pieces of equipment: the HP computer, iPad, and

Marantz are available. Recordings taken on newest models could potentially produce signals

clearer than those acquired in this study.

Group Differences

Another purpose of the present study was to determine if the quality of recordings for the

different devices depended on the participant differences such as age and gender identity. Due to

limitations in the adult sample, the researcher did not pursue the gender comparisons. Sample

size was small and because of a transmission issue, one of the male’s recordings were missing

from analyses, resulting in an unequal number of males and females. However, the researcher

did make comparisons across the adult and child groups. Peak clipping was one of the measures

employed in this study and the results revealed peak clipping only in the child samples. Peak

clipping occurs when the amplitude of a signal exceeds the level that recording device and

microphone can handle. Before making recordings, the researcher adjusted the recording levels

for each device to achieve optimal levels. Apparently, these adjustments were less accurate for

child speakers. Another measure used in this study was signal-to-noise ratio (SNR). Some

differences occurred across the adult and child speakers. In the reading passage task the SNRs

for adult and child speakers overlapped but the results for the child speakers were more variable,

based on standard errors of measurement and confidence intervals. For the word list task, the

iPad recordings of adult speakers yielded higher SNRs than the iPad recordings of child

speakers. For this task, the adult speakers had more variable results than the child speakers.

Finally, the conversational speech task yielded no clear differences in SNR between the adult

and child speakers for any of the recording devices.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 24

Environment

Much of the variability in the results of SNR analysis can be attributed to the difficulty of

controlling the recording environment. Even after instruction, child participants had a difficult

time sitting still and maintaining a constant distance from the microphones. Each session was not

recorded in the same therapy room of the speech and hearing clinic where the research took

place. Room acoustics could contribute to the signal and noise values of recordings. Speakers

produced sounds of greater intensity during the speech tasks than during the meter check,

making clipping possible. High frequency sounds, like fricatives, were produced with greater

intensity within the speech tasks. Also, natural rises in intensity occurred during the

conversational samples as many speakers showed enthusiasm in their responses. Clipping

resulted in a displayed signal value of 1 for the recordings in which it occurred, resulting in a

higher signal-to-noise ratio and raising the mean for the sample. Without clipping, average

values would have been slightly different, which may have altered comparison within the events

that featured clipping. These variables are similar to those that would be encountered when

recording during an evaluation, therapy session, or research study.

Future Research

Further investigation into optimal microphone pairings could provide higher quality for

all recording technologies. If head mounted microphones were used according to National Center

for Voice and Speech’s recommendations, and recordings of the same speech task on different

devices were taken individually, results may have been more consistent, providing a clearer

comparison. In this study, analyses focused on basic acoustic features. This may have

contributed to the lack of a clearly superior recording apparatus, as the study by Vogel and

Maruff (2008) suggested. Future research could investigate why the Marantz SNR values were

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 25

more consistent during the reading passage and word list tasks, and if this consistency were

necessary for more sophisticated acoustic analysis.

Conclusions

Transcription and reevaluation of speech samples is important to students and

professionals in the field of communication disorders for research, instruction, and evaluation.

Recordings of speech samples allow speech-language pathologists to focus on eliciting the

intended sample and maintaining the environment and interaction for the client. Professional

speech recorders intended for research are expensive and have few uses. More common

technologies, such as personal computers and iPads, are capable of recording speech samples and

can be used for a variety of purposes in both clinical and educational settings.

This study analyzed recordings taken by three different devices: a Marantz PMD 660

dedicated recorder, an HP personal computer with Audacity application, and a generation 4 iPad

with Voice Record Pro app paired with three omnidirectional, condenser lavalier microphones.

Recordings featured three speech tasks: a standard passage, word list, and conversational speech

sample, produced by adult and child participants. Following analysis for signal-to-noise ratio and

clipping using Praat software, it was concluded that none of the recording systems were clearly

superior to the others. Though data sets were varied, each technological device was capable of

producing high quality signals, indicating that readily available devices can be used for these

types of analyses.

Speech-language pathologists use laptop computers and tablets for a variety of functions,

including therapy activities. This study indicated that these devices are suitable for basic acoustic

analyses used to examine clients’ speech. Therefore, their use could be extended to include

recordings of speech samples, resulting in lower costs and greater efficiency for professionals.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 26

References

American Speech-Language Hearing Association (2016). Scope of practice in speech-language

pathology. American Speech-Language Hearing Association.

Barsties, B., & De Bodt, M. (2014). Assessment of voice quality: Current state-of-the-art. Auris

Nasus Larynx, 42, 183-188.

Boersma, P.l & Weenink, D. (2017). Praat: doing phonetics by computer [Computer

program]. Retrieved from http://www.praat.org/

Deliyski, D., Evans, M., & Shaw, H. (2004). Influence of data acquisition environment on

accuracy of acoustic voice quality measurements. Journal of Voice, 19(2), 176-186.

Howard, D.M., & Murphy, D.T. (2007). Voice science acoustics and recordings. San Diego, CA:

Plural Publishing.

Huckvale, M. (2009). Equipment for audio recording of speech. Retrieved from

http://www.phon.ucl.ac.uk/resource/audio/recording.html

International Phonetic Association (1999). Handbook of the International Phonetic Assocation:

A guide to the use of the International Phonetic Alphabet. Cambridge, UK: Cambridge

University Press.

Louko, L., & Edwards, M. (2001). Issues in collecting and transcribing speech samples. Topics

in Language Disorders, 21(4), 1-11.

Small, L. H. (2016). Fundamental of phonetics: A practical guide for students (3

ed.). Boston,

MA: Pearson.

Svec, J., & Granqvist, S. (2010). Guidelines for selecting microphones for human voice

production research. American Journal of Speech-Language Pathology, 19, 356-368.

van Lieshout, P. (2003). Praat short tutorial: A basic introduction. Retrieved from https://

web.stanford.edu/dept/linguistics/corpora/material/PRAAT_workshop_manual_v421.pdf

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 27

Vogel, A., & Maruff, P. (2008). Comparison of voice acquisition methodologies in speech

research. Behavior Research Methods, 40(4), 982-987.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 28

Appendix A

Informed Consent for Adult Participants

Project Title: A Comparison of Technologies for Recording Speech

Name of Investigator(s): Jacquelyn Knustrom

Invitation to Participate: You are invited to participate in a research project conducted through

the University of Northern Iowa. The University requires that you give your signed agreement to

participate in this project. The following information is provided to help you made an informed

decision about whether or not to participate

Nature and Purpose: The purpose of this research is to provide evidence as to whether or not

non-professional recording devices can produce recordings similar in quality to professional

recording devices and therefore be used as a cheaper, more readily available apparatus for

obtaining speech samples for instructional purposes. Acquiring quality speech samples plays a

vital role in transcription for Speech-Language Pathologists to accurately assess errors in

production of speech sounds.

Explanation of Procedures: You will be asked to respond to several background questions

before participating. This will include identifying your age, gender, hearing status, and history of

speech disorders. You will then read a word list, read a short passage, and engage in a

conversational narrative with the investigator. These speech samples will be recorded on three

separate devices so the quality of recordings can be compared. Your recording will be labeled by

your age and gender so the investigator can determine if those factors affect the quality of

recording. Your recording will be completed in a single session. Recordings will be stored

securely in the Roy Eblen Speech and Hearing Clinic following the conclusion of the study for

possible instructional use. Age and gender will be the only identifying information saved with

recordings.

Discomfort and Risks: Risks to participation are similar to those experienced in day-to-day life.

Benefits and Compensation: You will receive no direct benefits or compensation for your

participation.

Confidentiality: Information obtained during this study which could identify you will be kept

confidential. The summarized findings with no identifying information may be published in an

academic journal or presented at a scholarly conference. Recordings will be stored in a secure

area of the Roy Eblen Speech and Hearing Clinic on a password protected USB drive. In the

future recordings will be used only for academic research or instruction. Your name and

identification number will not be available with the recording following the conclusion of this

study.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 29

Right to Refuse or Withdraw: Your participation is completely voluntary. You are free to

withdraw from participation at any time or to choose not to participate at all, and by doing so,

you will not be penalized.

Questions: If you have questions about the study or desire information in the future regarding

your participation or the study generally, you can contact Jacquelyn Knustrom at 319-316-2034.

You can also contact the office of the IRB Administrator, University of Northern Iowa, at 319-

273-6148, for answers to questions about rights of research participants and the participant

review process.

Agreement:

I am fully aware of the nature and extent of my participation in this project as

stated above and the possible risks arising from it. I hereby agree to participate in

this project. I acknowledge that I have received a copy of this consent statement. I

am 18 years of age or older.

_________________________________ ____________________

(Signature of participant) (Date)

_________________________________

(Printed name of participant)

_________________________________ ____________________

(Signature of investigator) (Date)

_________________________________ ____________________

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 30

Appendix B

Informed Consent for Child Participants

Invitation to Participate: Your child has been invited to participate in a research project

conducted through the University of Northern Iowa. The University requires that you give your

signed agreement to allow your child to participate in this project. The following information is

provided to help you made an informed decision whether or not to participate.

Nature and Purpose: The purpose of this research is to provide evidence as to whether or not

non-professional recording devices can produce recordings similar in quality to professional

recording devices and therefore be used as a cheaper, more readily available apparatus for

obtaining speech samples for instructional purposes. Acquiring quality speech samples plays a

vital role in transcription for Speech-Language Pathologists to accurately assess errors in

production of speech sounds.

Explanation of Procedures: You will be asked to respond to several background questions on

behalf of your child before he or she participates. This will include identifying his or her age,

gender, hearing status, and history of speech disorders. Your child will then read or repeat a

word list, read or repeat a short passage, and engage in a conversational narrative with the

investigator. These speech samples will be recorded on three separate devices so the quality of

recordings can be compared. Your child’s recording will be labeled by his or her age and gender

so the investigator can determine if those factors affect the quality of recording. Your child’s

recording will be completed in a single session. Recordings will be stored securely in the Roy

Eblen Speech and Hearing Clinic following the conclusion of the study for possible instructional

use or further research. Age and gender will be the only identifying information saved with

recordings.

Discomfort and Risks: Risks to participation are similar to those experienced in day-to-day life

Benefits: Your child will receive no direct benefits or compensation for his or her participation.

Confidentiality: Information obtained during this study which could identify your child will be

kept strictly confidential. The summarized findings with no identifying information may be

published in an academic journal or presented at a scholarly conference. Recordings will be

stored in a secure area of the Roy Eblen Speech and Hearing Clinic on a password protected

USB drive. In the future recordings will be used only for academic research or instruction. Your

child’s name and identification number will not be available with the recording following the

conclusion of this study.

Right to Refuse or Withdraw: Your child’s participation is completely voluntary. He or she is

free to withdraw from participation at any time or to choose not to participate at all, and by doing

so, your child will not be penalized.

Questions: If you have questions about the study or desire information in the future regarding

your child’s participation or the study generally, you can contact Jacquelyn Knustrom at 319-

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 31

316-2034. You can also contact the office of the Human Participants Coordinator, University of

Northern Iowa, at 319-273-6148, for answers to questions about rights of research participants

and the participant review process.

Agreement:

I am fully aware of the nature and extent of my child’s participation in this project

as stated above and the possible risks arising from it. I hereby agree to allow my

son/daughter to participate in this project. I have received a copy of this form.

_________________________________ ____________________

(Signature of parent/legal guardian) (Date)

_________________________________

(Printed name of parent/legal guardian)

_________________________________

(Printed name of child participant)

_________________________________ ____________________

(Signature of investigator) (Date)

_________________________________ ____________________

(Signature of instructor/advisor) (Date)

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 32

Appendix C

Grandfather Passage

You wish to know all about my grandfather. Well, he is nearly 93 years old, yet he still thinks as

swiftly as ever. He dresses himself in an old black frock coat, usually several buttons missing.

A long beard clings to his chin, giving those who observe him a pronounced feeling of the

utmost respect. When he speaks, his voice is just a bit cracked and quivers a bit. Twice each

day he plays skillfully and with zest upon a small organ. Except in the winter when the snow or

ice prevents, he slowly takes a short walk in the open air each day. We have often urged him to

walk more and smoke less, but he always answers, “Banana oil!” Grandfather likes to be

modern in his language.

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 33

Appendix D

Caterpillar Passage

Do you like amusement parks? Well, I sure do. To amuse myself, I went twice last spring. My

most MEMORABLE moment was riding on the Caterpillar, which is a gigantic roller coaster

high above the ground. When I saw how high the Caterpillar rose into the bright blue sky I knew

it was for me. After waiting in line for thirty minutes, I made it to the front where the man

measured my height to see if I was tall enough. I gave the man my coins, asked for change, and

jumped on the cart. Tick, tick, tick, the Caterpillar climbed slowly up the tracks. It went SO high

I could see the parking lot. Boy was I SCARED! I thought to myself, “There’s no turning back

now.” People were so scared they screamed as we swiftly zoomed fast, fast, and faster along the

tracks. As quickly as it started, the Caterpillar came to a stop. Unfortunately, it was time to pack

the car and drive home. That night I dreamt of the wild ride on the Caterpillar. Taking a trip to

the amusement park and riding on the Caterpillar was my MOST memorable moment ever!

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 34

Appendix E

Word List

choice

swing

those

eat

witch

kid

crab

thunder

hand

teacher

family

globe

fishing

home

reward

star

yellow

zipper

spider

cow

mouse

oatmeal

taught

move

cat

pajamas

sled

leaf

present

wife

toy

shoe

vacuum

mother

pencil

toe

window

barn

blue

eye

nose

breathe

thank

goat

baby

crackers

jumping

eight

leash

smoke

off

clown

fish

itch

tiger

sky

train

bet

five

rope

weigh

beige

look

muscle

alien

behind

flag

twins

mean

chair

about

hour

mask

plant

teeth

oil

ask

orange

tame

quick

tree

television

green

umbrella

carrot

duck

bathtub

end

oops

telephone

buy

snake

A COMPARISON OF TECHNOLOGIES FOR RECORING SPEECH 35