About the presenters:
Mike Susca is a doctoral student at the University of Nebraska-Lincoln. Prior to enrolling in doctoral studies, he was a practicing speech-language pathologist for the past 21 years in San Diego, California. He has considerable clinical experience with clients of all ages in multiple settings and multiple communication disorders. He is a fluency specialist and will focus his Ph.D. program in the area of fluency disorders.

E. Charles Healey is a professor of speech-language pathology in the Department of Special Education and Communication Disorders at the University of Nebraska-Lincoln. He has been a faculty member at the University of Nebraska for the past 21 years. He has published and presented numerous papers on fluency disorders in children and adults. He is an ASHA Fellow.


Is Stuttering the Only Thing People Listen To?

by Mike Susca and E. Charles Healey
from Nebraska, USA

INTRODUCTION

Across several decades, researchers have been interested in listener judgments of stuttering behavior. Listener reactions to stuttering have played an important role in shaping our clinical perspectives about stuttering. Assessment of listener perceptions has generally been on singular dimensions such as word recall, story comprehension, or general severity of stuttering (Cypress, Hezel, Rossi, and Adams, 1984; Hulit, 1976). Affecting those perceptions has been a focus on the number and duration of repetitions, prolongations, and blocks (Leach, Wolfolk, Fucci and Gonzales, 1995 & 1997).

These past studies have generated important findings regarding listener's ability to a) identify stuttering within a speech sample, b) decide whether a sample of speech is from a person who stutters or a normally-fluent person and c) comprehend the content of a message that contains disfluent speech. However, all of these past studies have investigated a rather narrow perspective on listener reactions to stuttering. We believe it is important to consider broader questions related to listener response to stuttering so we can gain a better understanding of how listeners react to and perceive various forms and frequencies of stuttering. For example, how does the form and frequency of stuttering influence a listener's perception of the competence of the speaker or whether the person is easy to understand? Do listeners react differently to "mild" than to "severe" stuttering and if so, how? Are there any features of speech that listeners' attend to when they hear a fluent speech sample from a person who stutters? Answers to these questions could have direct application to the clinical management of stuttering in that the client and clinician can make a more informed decision about the quality of speech at various stages of treatment.

The purpose of this study was to explore a number of broader issues and questions about listener perceptions of stuttering than has been studied in past research. Specifically, we sought to discover if there would be differential effects on listener judgments of speaker performance to increases and decreases in the number of stuttering moments and changes in prosodic features contained in the speech signal.

METHODOLOGY

A speech sample was obtained from a 27-year-old male who stuttered who was being seen for an evaluation of his stuttering. The sample was taken from the client's first oral reading of the Rainbow Passage during the evaluation. Later analysis of the speech sample by five independent judges revealed reliable judgments of stuttering on 10% of the words in the passage. The analog recording of the passage was then digitized in it's entirety using a speech synthesis and analysis microcomputer program (Soundforge, 1998). This digitized "core" sample was copied four more times to create four separate files on the computer. Four samples of stuttering were developed by either increasing or decreasing the number of stuttering events. Removing and adding stuttering events to the 10% sample took considerable time and effort to insure that any manipulation of the waveform did not produce artifacts in the signal that a listener would hear.

Construction of the Speech Samples

The first modified sample was modified by adding 5% more stuttering to the core sample . To increase the stuttering frequency, we electronically copied part-word repetitions and sound prolongations that occurred. The form and duration of the new stuttered events,which were placed into new locations within the passage, were identical to the original stuttering events. The second modified sample was constructed by removing half of the part-word repetitions, whole-word repetitions, and audible sound prolongations in the core sample. This resulted in a 5% sample. Elimination of stuttering events involved carefully deleting and reshaping the waveform of the sound or word stuttered.

A third modified sample involved eliminating all stuttering events from the core sample such that a 0% sample was created. All other aspects of the speech sample remained the same (i.e. speech rate and pauses). In order to determine objectively that the 0% sample did not contain any perceptible stuttering, we asked a small group of speech-language pathologists to listen and evaluate the sample. They commented that stuttering events were absent but the speaker seemed to have a slow speech rate and had occasional episodes of glottal fry. Consequently, we developed a fourth sample for this experiment. In this sample, we reduced the frequency and duration of pause times, eliminated extraneous glottal pulses, and lengthened the duration of onset and offset times on words visible in the waveform. This sample was called the "treated" sample. One final sample from a normal speaker was constructed in order to compare that sample to the "treated" sample. A 24-year-old male without any history of a speech or language disorder read the first paragraph of the Rainbow passage. The sample was digitized but not modified in any form. This sample was called the "normal" sample.

Procedure

We obtained 60 listeners for this experiment (13 males and 47 females between the ages of 19 and 52). Ten individual listeners with limited exposure or experience with persons who stutter listened to each sample. These listeners were not informed as to what they would hear in the speech sample. Each listener passed a hearing screening test and heard only one sample once to eliminate any bias that could occur by listening to more than one sample (Sander, 1965).

After hearing a sample, each listener was given a list of four statements. Their task was to rate the speech sample on a 7-point Likert scale from strongly disagree (1) to strongly agree (7). The statements were:

  1. This person is a competent speaker.
  2. This person is a fluent speaker.
  3. This person read the story easily.
  4. I felt comfortable listening to this person.
Following the Likert ratings, each listener was asked four general questions about the speech sample they heard. Listener's verbal responses were tape recorded and later, transcribed verbatim. The four questions were:
  1. How would you describe this person's speech?
  2. What contributed to your decision regarding how comfortable you feel listening to this person?
  3. What were some of the features of his speech that caught your attention?
  4. What contributed to your response regarding how easy/difficult it was to understand the story told on the tape?
We analyzed the transcripts by assessing whether each statement a listener made about the speech sample was either a positive or negative comment. A positive comment was defined as a complementary description or statement about normally perceived attributes of a speaker. A negative comment was defined as derogatory description or statement about the attributes of a speaker.

WHAT DID WE FIND?

Results are divided into Likert scale ratings and the number of positive/negative responses for each speech sample.

General Likert Scale Ratings Results:

The mean Likert scale ratings are depicted in Figure 1.

From the graph it can be seen that the Likert scale ratings across all five statements decreased as the frequency of stuttering increased. Interestingly, the 0% sample (no stuttering) was rated lower than the normal or treated samples. Also, the normal and treated samples were virtually equivalent. Recall that the only difference between the 0% and treated samples was the modification of prosodic elements within the samples.

The Likert scale data also showed that listeners of the normal sample rated the speaker as more competent then listeners of the 15% percent sample. Also, the normal and treated samples were perceived as being significantly more fluent than the 5%, 10% and 15% samples. The ratings from listeners of the normal and treated sample showed that their perception of the speaker's ease of reading the story was significantly higher than the 0%, 5%, 10% or 15% speech samples. Moreover, listeners of the treated sample perceived that the sample was significantly more comfortable to listen to than the 10% and 15% samples.

Positive and Negative Statements

Figure 2 shows the average percentage of positive and negative comments listeners made to all questions in each speech sample.

From this figure you can see there is a gradual decrease in the percentage of positive comments as the sample of stuttering increases. The exact opposite pattern is shown for the percentage of negative comments. Note that the treated sample is almost a mirror image (of positive and negative comments) to the zero percent sample -- even though both samples did not contain any stuttering. Specifically, listeners commented more positively about the normal and treated samples than all other samples in terms of listening comfort, ease of understanding the story, and features of the speech that caught their attention.

In contrast to the positive comments made about the samples, listeners expressed more overall negative perceptions about the samples containing stuttering. Listeners made specific negative comments directed at the speaker's personality ("negative", "frustrated"), comfort ("not comfortable"), confidence ("nervous", "unsure"), competence ("not competent"), believability ("confused", "had to think about things"), or distracting speaking style ("had to revise", "no excitement", "struggled", "incomplete"). The number of negative comments varied across samples but clearly, as little as 5% stuttering was disconcerting to listeners. It is interesting that the comments associated with the samples of stuttering dealt with aspects of the speaker other than repetitions, prolongations, or blocks.

Listeners expressed more difficulty understanding the story as the sample of stuttering increased. The primary contributor to this pattern was that increasingly too much cognitive effort was required with each increased sample of disfluency. Essentially, the listener's train of thought was disrupted by the stuttering and they had to listen carefully to the message when frequency of stuttering within the sample increased.

WHAT DO OUR RESULTS SUGGEST?

This study revealed three major points regarding listener perceptions of various degrees of stuttering and fluency:

Listeners appear to react negatively with gradual increases in perceived stuttering. However, it is interesting to note that samples of speech that did not contain stuttering can elicit negative ratings and reactions if the speaker changes expected prosodic features of speech (e.g., frequency and extent of pauses, glottal fry, and sudden voice onset and offset times. This finding suggests that stuttering treatment programs aimed at establish completely nonstuttered speech need to consider the quality of the speech that occurs even though repetitions and prolongations are absent. In other words, our data confirm a long-held clinical notion that 100% fluency doesn't necessary mean the speech will be perceived as normal. Improving the prosodic features of speech (i.e. making it more natural sounding) results in more positive comments from listeners.

The results of this study also show that listener perceptions of stuttered and fluent speech go beyond the number and duration of stuttered moments. Issues such as overall speaker fluency, speaker competency, perceived ease in reading a passage, listener's comfort listening, and general ease of understanding a story are other dimensions affecting listener perceptions of speech communication.

The qualitative positive and negative responses related to listeners perceptions of the speech samples provided greater insights into how people perceive various levels of stuttering than the Likert scale data. From a clinical perspective, it may be helpful for clinicians to have a group of listeners evaluate and comment on samples of speech obtained from the adult or child who stutters. These responses could be used to determine the perceived quality and competence of the client's speech at various stages of treatment.

REFERENCES

Cyprus, S., Hezel, R.T., Rossi, D. & Adams, M.R. (1984). Effects of Simulated Stuttering on Listener Recall, J. of Fluency Disorders, Vol.17, 191-197.

Hulit, L.M., (1976), Effects of Nonfluencies on Comprehension , Perceptual and Motor Skills, Vol. 42, 1119-1122.

Leach, Edwin A., Wolfolk, William B., Fucci, Donald, & Gonzales, Maria Diana. (1995). Simulations of Stuttered Speech: Numbers and Types of Dysfluencies, Perceptual and Motor Skills, Vol. 81, 367-370.

Leach, Edwin A., Wolfolk, William B., Fucci, Donald, & Gonzales, Maria Diana (1997). An Investigation of Simulated Stuttering by Normal Speakers, American Speech-Language-Hearing Association Convention, Boston, Massachusetts.

Sander, E.K., (1965). Comments on Investigating Listener Reaction To Speech Disfluency, J. of Speech and Hearing Disorders, Vol. 30, N0.2, 159-165.


September 9, 1999