By Bobby E. Waldrup, Jeffrey Michelman, and Maxine L. Person
A Promising Technology Not Ready for Prime Time
Voice recognition software (VRS) allows computer users to control applications with voice commands instead of the traditional keyboard and mouse. The technology is being heralded as the next generation of human-computer interface, promising to make computer applications more efficient and user-friendly.
The authors introduce the three major off-the-shelf VRS packages, discuss possible uses for CPAs, and chronicle their difficulties in using the technology and adapting to its requirements and shortcomings. Those hoping for immediate increased efficiency in all aspects of usage will likely be disappointed; VRS delivered on certain promises but fell far short of the authors’ expectations in others. Prospective users are warned that off-the-shelf packages are not necessarily ready to fulfill all the needs of the demanding CPA.
Now that Y2K has come and gone, CPAs can consider investing in new computer technologies, searching for software that offers innovative new capabilities. Voice recognition software (VRS) is a relatively new technology with a great deal of promise. VRS allows professionals to control computer applications with voice commands in lieu of the traditional keyboard and mouse. The ultimate promise of the technology is to allow humans to interact with computers more naturally and efficiently; however, the future is a long way off. Current VRS offerings provide significant benefits, but not without accompanying limitations and drawbacks, especially for technical professionals.
VRS provides for the two broad functions of voice-activated dictation and macro-level control over applications. Dictation is the most commonly used function. With a headset attached to the computer or a digital voice data-collector similar to the hand-held tape recorder, users can directly record unlimited dictation into popular word-processing programs such as Microsoft Word and Corel WordPerfect.
In the second most commonly used VRS function, voice commands—rather than a mouse or keyboard—are used to operate applications. The user can also voice-activate applications to save, print, and move files and perform mathematical functions in spreadsheet and database applications.
Vendors. The majority of the VRS market share is controlled by three vendors: IBM (ViaVoice), Lernout & Hauspie (Voice Xpress), and Dragon Systems (Dragon Naturally Speaking). [At the time of writing, Lernout & Hauspie had acquired Dragon Systems as a wholly owned subsidiary. Although L&H Senior Product Director Bill DeStaefanis publicly announced that both brands will continue to be sold, prospective adopters should consider the effect the merger may have on future product offerings.
The three vendors make similar claims about the capabilities of their respective brands. All indicate spelling and grammar accuracy upwards of 95% for dictation applications. They each state that initial voice recognition will require 12–15 minutes training time but that longer use will allow for greater accuracy. According to vendor marketing claims, the software should be capable of integration with other Windows-based (and, in the case of the IBM brand, Macintosh-based) business applications.
The purpose of this article is to share the authors’ findings with respect to these vendor claims. We argue that accountants should use caution before adopting this new technology for intensive professional use.
Current Trends in Practice
All areas of accounting practice have been, to some degree, revolutionized by automation. This can best be illustrated by the explosion of prepackaged PC-based software that can assist in tax preparation and general ledger, database, spreadsheet, and work-paper delivery systems. These innovations are in addition to commonly used office-based software for word processing, e-mail, and presentation. Accounting applications, while producing significant efficiencies in a professional office, still must generally be operated in a stand-alone fashion. The efficiencies gained by computer technologies can be enhanced by the user’s ability to move data fluidly among different packages.
Practitioners may be tempted to adopt VRS as a means of efficiently toggling between various accounting and office applications. VRS vendor marketing efforts portray office professionals moving effortlessly among a wide variety of these applications using natural language technology (NLT) without resorting to keyboard or mouse commands.
The three dominant VRS packages mentioned above were tested to determine their applicability to various accounting related tasks. These packages share the following commonly advertised features:
To facilitate analysis, each of these software packages was installed on a Windows desktop PC. The computer’s hardware system significantly exceeded the minimum requirements necessary according to all vendor specifications. The investigators analyzed the performance of the VRS packages as a group based upon the following criteria (See Exhibit 1):
Each of the packages was evaluated on an A–F scale by two separate evaluators. These grades were then averaged to produce an overall technology grade, rather than a comparative ranking of the individual software packages, which generally performed similarly (See Exhibit 2). The grade represents an evaluation of the software’s performance as it applies to the typical CPA. This grading scale is inherently subjective and is meant only to serve as broad guidance (See Sidebar). Recommendations of specific software packages are not implicit in the evaluations.
Ease of installation, learning curve, ease of use. The investigation found that each of the packages installs quickly, and learning to get around the programs is easy and relatively painless. Training protocols are simple, and a PC-literate professional will have any of the packages running within an hour. One concern is that the programs had difficulty learning specialized vocabulary, such as proper nouns and specific accounting terms.
Software performance: dictation and command functions. VR technology excels at quickly and accurately recognizing common speech patterns. Each VRS package touts dictation accuracy upwards of 95% using commonly recognized vocabulary. However, once the user deviates from even the most basic of vocabulary, the technology begins to falter. Concerns have to do with three basic failures: Pervasive background interference during use, Dictation into non-word-processing applications, and Dictation of specialized vocabulary.
Ambient background noise during dictation significantly reduces the accuracy of each of the software packages. Heating and cooling fans sometimes cause significant error rates in dictation. If a telephone rings or an intercom buzzes, each of the software packages will continue to respond, typing gibberish. In addition, if the user is hoarse or congested, the accuracy of dictation can be severely hampered.
Each of the packages allows for dictation into proprietary notepad programs, which can be translated into any of the major word-processing packages. Each also allows for dictation directly into other application programs, such as databases, e-mail, and spreadsheets. We found this dictation application to falter significantly with regard to accuracy and speed. For example, one investigator with a southern accent found that dictating the words “effort” and “forth” into an Excel spreadsheet would repeatedly cause the cursor to jump to the cell “F4.” Another investigator had difficulty with the command functions; when commanded to “delete that,” the software consistently typed “believe that.”
The investigators also found that specializing the vocabulary of these VRS packages is not as easy as advertised. Dictation of special terminology such as SAS, FASB, and client corporate and individual names quickly becomes burdensome. Several proper names were never correctly learned by the packages, despite concentrated and repeated specialized training that is far beyond the patience and available time of most CPAs. While each of the applications allows for numerical and dollar-denominated nouns (e.g., $1,232.45), vendor-specific pronunciations of all numerical amounts must be used for the software to comprehend.
The promise of VR technology to create a voice-controlled command environment in Windows-based applications is disappointing. Using voice commands in lieu of the mouse is an arduous process. Often, out of sheer aggravation, the investigators would simply reach for the mouse to complete a simple command such as printing. While it does appear that the packages can substitute voice commands for practically all mouse and keyboard functions, in practice this feature would only appear to be of benefit to individuals that could not use or encountered extreme physical difficulty when operating a traditional mouse and keyboard.
The VRS report card (Exhibit 2) indicates a wide variability of performance in different functional areas. Although there was slight variation in performance among the three products, this did not impact the overall assessment. Perhaps the most surprising result of this study is that the overall evaluation of VRS software for CPAs’ tasks is between a C and C-. Most CPAs would hardly choose to hire staff with this type of performance record; therefore, it is unlikely that they will choose to adopt technology at this level of performance.
Current Practice Needs Using New Technology
As currently packaged, the only real strength of VRS technology lies in simple dictation. This is unfortunate for CPAs, because professional usage would not generally focus upon these basic tasks. Currently, the needs of CPAs revolve around e-mail applications, control of quantitative office software (i.e., spreadsheets and databases), client-centered work-paper production, and control of various proprietary professional accounting packages. Even allowing an optimistic 95% accuracy rate in all applications would leave open the possibility of serious error.
Proprietary accounting-specific VRS is both possible and probable in the near future. Specialized software can be written effectively for profession-specific applications, as demonstrated in the numerous packages available for health care tasks such as medical transcriptions, patient billing records, and pharmaceutical prescriptions. However, this kind of specialized software is not currently available for the accounting profession; off-the-shelf applications such as those examined above are the only available alternative for CPAs hoping to adopt VR technology.
Room for Improvement
In order to examine accountant needs for high-quality VRS, the authors interviewed eight practicing CPAs and a tax attorney. All nine had some knowledge of the VRS but none had ever tried it. The consensus was that anything which could improve the efficiency of various aspects of their work would be welcome (The results are presented in Exhibit 3). Perhaps the most pervasive issue for accountants, their employees, and their clients is the possibility of suffering from repetitive stress injuries, such as carpal tunnel syndrome. Hence any technology that could reduce the incidence of these debilitating injuries would be welcomed.
The group generally agreed that the keyboard and mouse have become barriers to data entry through either lack of typing skills, limitations of time and place, or the inability to foster an environment for the dialogue necessary for Internet or tax research. For these reasons, the group was excited about the possibilities of VRS—if its accuracy were improved, its integration with applications were enhanced, and it could facilitate better research capabilities through knowledge management. Finally, several interviewees noted that VRS needed to be used with caution at client sites due to the confidential nature of the data being entered.
VRS is a relatively new technology. Off-the-shelf packages can be initialized, by a technologically savvy user, in under an hour and be trained, with a slight learning curve, to handle verbal dictation with an accuracy upwards of 95%. However, CPAs need significantly more complex forms of automation. Currently available packages will not meet the expectations of the typical accountant. A certain level of professional skepticism is warranted; the authors recommend that CPAs delay adoption of VRS until it is further developed for non-word-processing applications by the major vendors or until tested, accounting-specific versions of the software are available.
In looking toward the future, we feel that it is particularly important to put the state of VR technology into perspective. During the early 1980s it would have been hard to find a significant number of CPAs who would champion the value of electronic spreadsheets or, for that matter, PC technology in general. Furthermore, many of us welcomed the AICPA partnership with the Grammatik software in the early 1990s, which made grammar checking much easier. We clearly envision the development of an accounting-based VRS dictionary that would provide improved functionality similar to those applications currently available to the medical and legal professions.
The long-term future development of VR technology hinges upon developing a neural network technology that mimics the way the brain interprets speech. Neural networks use whole word analysis rather than the current approach, which segments speech patterns into smaller chunks. We believe that once computer science advances to a more brain-like interpretation matrix, VR technology will indeed become the new standard of data entry. Thus, even though we recommend against early adoption of VR technology, we strongly recommend that the accounting profession continue to follow this emerging standard.
CPAs might consider the day when VR technology facilitates scripted client interviews, whereby clients can easily prepare auditor checklists that are nonintrusive, yet accurate. Great strides in VR technology have already been achieved in other professional disciplines. Although they have yet to achieve the productivity and accuracy rates that CPAs require, advances in the technology will have a profound effect on the profession.
The CPA Journal is broadly recognized as an outstanding, technical-refereed publication aimed at public practitioners, management, educators, and other accounting professionals. It is edited by CPAs for CPAs. Our goal is to provide CPAs and other accounting professionals with the information and news to enable them to be successful accountants, managers, and executives in today's practice environments.
©2009 The New York State Society of CPAs. Legal Notices
Visit the new cpajournal.com.