Comparing real-time and transcript-based techniques for measuring stuttering

Abstract

The purpose of this study was to compare results obtained from two different techniques for counting speech disfluencies. Fifty audio-videotaped speech samples were analyzed using: (a) a transcript-based technique designed to evaluate speech (dis)fluency in the context of a speaker’s conveyed message and (b) a real-time technique designed to rapidly determine the frequency of various types of speech disfluencies in conversational speech. Results obtained using these two techniques were quite similar, though there were some consistent and predictable differences associated with the presence of brief sound prolongations and more complicated disfluency clusters. Further analysis revealed that the two measurement techniques resulted in substantially different severity ratings for only two of the 50 speech samples. Findings provide support for a comprehensive measurement strategy in which clinicians utilize a transcript-based approach when more detailed information is needed (e.g., during diagnostic evaluations) and a real-time approach for documenting on-going changes in clients’ speech behaviors (e.g., during treatment).

Introduction

In recent years, there has been considerable debate regarding the measurement of stuttering behaviors. One of the most prominent topics in the literature has been measurement reliability Cordes 1994, Cordes & Ingham 1994a, Curlee 1981, Kully & Boberg 1988, MacDonald & Martin 1973, Tuthill 1946, Young 1975, Young 1984; however, there are a number of other important issues in need of further study (Yaruss, 1997a). For example, recent research has examined whether it is appropriate to count instances of disfluency or instances of stuttering (e.g., Conture 1990, Cordes & Ingham 1996a, Costello & Ingham 1984, Ham 1989), whether such counts should be based on the number of words or number of syllables produced Andrews & Ingham 1971, Brundage & Bernstein Ratner 1989, Ham 1986, or whether counts should be based on the instances of speech disruptions or the time intervals containing disruptions Cordes & Ingham 1994b, Cordes & Ingham 1994c, Cordes & Ingham 1995a, Cordes & Ingham 1996b, Cordes et al. 1992, Ingham, Cordes, & Finn 1993, Ingham, Cordes, & Gow 1993. It seems likely that clinicians’ varying decisions regarding these variables may contribute to the finding that different clinicians obtain different results when measuring stuttering Kully & Boberg 1988, Cordes & Ingham 1995a.

In addition to these topics, it seems reasonable to assume that there may also be important differences associated with the $\text{type}$ of measurement technique a clinician selects. Some methods for measuring stuttering emphasize a more comprehensive analysis of speech (dis)fluency based on a detailed verbatim transcript of a speech sample (e.g., Campbell & Hill 1987, Rustin, Botterill, & Kelman 1996), while others emphasize a more rapid but less detailed analysis of the production of speech disfluencies based on real-time (or “on-line,” after Conture, 1990) counting and categorizing of speech disfluencies (e.g., Conture 1990, Conture & Yaruss 1993, Riley 1994, Yaruss 1998). Certainly, each technique has its advantages and disadvantages. Transcript-based techniques provide more information than can readily be obtained through real-time measures, including more detailed analyses of how a client’s speech and language abilities relate to the production of speech disfluencies. Transcript-based techniques also provide greater opportunities to assess qualitative aspects of speech disfluencies, such as audible and visible tension. Unfortunately, transcript-based analyses are also quite time consuming, so it is generally not feasible to conduct such detailed measurements on a regular basis (e.g., for documenting changes in a client’s progress throughout treatment). Real-time techniques, on the other hand, are much faster to complete, so they provide a method for collecting the objective data necessary to document changes in a client’s stuttering behaviors without requiring a large time commitment. Still, the amount of detail that can be assessed with real-time techniques is somewhat limited. Based on the strengths and weaknesses of these two approaches, then, one seemingly reasonable measurement strategy would be to utilize transcript-based methods when more detailed data are necessary, such as during a diagnostic evaluation, and real-time methods when it is less feasible to devote a large amount of time to data collection, such as during treatment (Yaruss, 1997a).

Unfortunately, it is not presently clear how the results of real-time and transcript-based measures relate to one another. Given the concerns regarding the reliability of stuttering measurements noted above, it seems reasonable to assume that there may be differences in the results obtained using these different techniques. For example, real-time analyses require rapid judgments of speech behaviors, so subtle behaviors may be less likely to be identified. Conversely, there may be a tendency to “overanalyze” subtle behaviors with transcript-based procedures, when a videotaped speech sample can be viewed repeatedly or in slow motion. As a result, it is not clear whether data obtained using these two techniques can be reasonably compared in a strategy such as that proposed above. Accordingly, the purpose of this study was to examine similarities and differences in the frequency and types of speech disfluencies obtained using a transcript-based and a real-time analysis in order to determine whether these two types of measurement approaches can be combined in a comprehensive strategy for measuring the speech disfluency behaviors of individuals who stutter.