BBI Contributing Writer

In this brave new world where technology is always changing, society is becoming more dependent upon software. One emerging technology undergoing remarkable changes over the last few years is voice recognition. Voice recognition (VR) is an evolving technology that is beginning to mature thanks to faster computer chips, cheaper physical memory, improved speech engines, and better microphones. Voice recognition offers great potential benefits in the medical field, telephone industry, for large corporations, and handicapped individuals as well.

Chances are you have had contact with speech recognition already. Remember the last time you called information for a number and got a computer rather than a person? That is just one example of how speech recognition is beginning to infiltrate our society today. Where is speech recognition headed in the medical industry? What interest is there for voice products and in what departments is the need greatest? Who are the major players in the voice market now, and what advances are on the horizon?

The technology for voice recognition has been evolving since the mid-1950s, but fell out of favor in the 1960s, as it never achieved the accuracy required for useful applications. There was renewed interest in this field in the early 1970s, especially in the artificial intelligence community. But voice recognition did not springboard into the consumer electronics market until 1993 with the launch of IBM's (Armonk, New York) Personal Dictation System. In 1996, IBM introduced a real-time continuous voice recognition medical application, Medspeak/Radiology, and the stage was set for practical medical applications to evolve.

To gauge the growing interest in voice recognition applications within the medical community, consider the Healthcare Information and Management Systems Society's (HIMSS; Chicago, Illinois) 10th annual Leadership Survey. Completed in 1999, it reported that 17% of provider respondents planned on implementing a speech recognition system within one year. By the time this year's HIMSS survey was conducted, an eye-popping 46% of provider respondents reported planning to implement a VR system within the next two years. The 46% of providers who plan on implementing a system by the end of 2002 do not include the 20% of respondents who already have an installed voice recognition system, suggesting that VR technology is ready to emerge in the medical sector.

Hospital applications include radiology, pathology and emergency departments where the technology reduces turnaround time and the costs of the dictation/transcription process. By making workflow more efficient, VR helps to lessen the agonizing wait time a patient experiences. As voice recognition expands and gets "smarter," the demand for applications will increase. In the next five years, voice-driven electronic medical record (EMR) systems and CTP & DRG coding applications will become practical.

L&H's rise – and big fall

But the biggest story in voice recognition over the last year or so has not been the advance in technology, but rather the fall from grace of VR industry kingpin Lernout & Hauspie (L&H; Ieper, Belgium/Burlington, Massachusetts).

Lernout & Hauspie had become the global leader in speech and language technology through its own accomplishments and the aggressive acquisition of Kurzweil (Boston, Massachusetts) and Dragon Systems (Newton, Massachusetts). L&H was given instant credibility in the industry in 1997 when Microsoft (Redmond, Washington) invested $45 million in the Belgian company. Since then, Microsoft and Intel (Santa Clara, California) have invested millions more. L&H looked to be on its way up, but fraud and mismanagement led to its demise.

In May 2000, Lernout & Hauspie announced that it had finalized the acquisition of Dictaphone (Stratford, Connecticut). On paper, the acquisition looked attractive and there was a logical product "fit." L&H was using Dictaphone to expand its customer base in the medical dictation field, as Dictaphone's health care market assets included 5,000 medical industry clients worldwide, with a heavy concentration of those located within the U.S. Dictaphone had an excellent reputation for good service, and at the time had 400,000 physicians in U.S. using their transcription systems for clinical record processing. The purchase of Dictaphone also meant that over half of the Belgian upstart's business was now in the U.S. and subject to Securities & Exchange Commission (SEC) regulation. Prior to this purchase, L&H was disclosing the minimum information required by the SEC for foreign companies, but now would have to begin reporting in much greater detail. Lernout & Hauspie also assumed $425 million of Dictaphone's long-term debt.

In June 2000, L&H finalized its acquisition of Dragon Systems, a formidable competitor with award-winning speech recognition software. In 1997, Dragon Systems had introduced the world's first general-purpose, large-vocabulary, continuous-speech recognition product, Dragon NaturallySpeaking. The company offered specialized, vertical market vocabularies for medical applications.

L&H believed acquiring Dragon would help it penetrate new market arenas such as hand-held and mobile devices, and strengthen its hold in health care, where Dragon Systems arguably had a superior product offering. As with the Dictaphone deal, this purchase had risks for L&H. Even though Dragon had one of the superior voice recognition products, it was hampered by its own lack of marketing skill. L&H was going out on a limb again by acquiring a company that had posted a net loss of approximately $22 million for fiscal 1999.

By this time, L&H was beginning to lose steam. Even though it was a giant in the VR industry with many distribution channels, a large share of the medical market and few competitors left that it had not bought, the company was being hampered by bad management decisions. In the past, it had been hard for analysts to get a true perspective of the company's performance because of L&H's sparse SEC filings. Compounding this was the rapid-fire rate at which L&H had been acquiring companies. As more financial data was issued, it became clear to industry insiders that the numbers were not adding up.

Hard data supplants the whispers

Analysts were now starting to add hard data to the whispers they had been hearing for years. A reported jump in sales in the Korean subsidiary of L&H from about $100,000 in the first quarter of 1999 to some $59 million in the same quarter of 2000 was startling, considering that sales everywhere else in the world at the time were decreasing for Lernout.

In early August of last year, things finally came to a head. The Wall Street Journal reported that it had found discrepancies in the Korean sales figures. Bastiaens had earlier volunteered the names and approximate sales numbers for 30 of L&H's Korean customers. The WSJ blew holes all through the list, noting that three of the 30 were not even customers, while others reported that the sales figures cited by L&H were significantly higher than what they had actually paid for their VR systems. Then-CEO Gaston Bastiaens rushed to refloat his sinking ship by ordering a special audit by KPMG and flying market analysts to Korea to "see for themselves," but this did little to calm shaky investors.

In late August, the company announced that John Duerden, then president and CEO of L&H's Health Care Solutions Group and formerly chairman and CEO of Dictaphone, would be replacing Bastiaens as CEO of L&H. Before Duerden could even get a "To do" list together, more bad news flooded in. As a result of irregularities found in the special audit, the company announced that it would be restating its financials for 1998, 1999 and the first half of 2000. It also conceded that revenues for the 3Q00 would be at least $40 million below its previously stated forecast range of $165 million to $185 million. Not only was it bad news for L&H, but terrible news for KPMG, whose performance was shown to be sloppy and incompetent in financial reviews by outside auditors.

The special audit also showed a shortfall on the balance sheet of the Korean subsidiary. The story is that L&H purchased Bumil Information and Communication Co. Ltd. from Ju-Chul Seo for $25 million in September 1999. The purchase agreement provided for an earn-out of $25 million to be paid to Seo based on results in the Korean subsidiary between the acquisition date and Dec. 31, 2000. To inflate the numbers, the subsidiary made it appear that it had received roughly $108 million in orders that it did not actually have, doing so through an elaborate banking scheme.

In January of this year, Philippe Bodson was appointed the new CEO, replacing Duerden. Bodson previously was CEO of Tractebel, a Belgian-based utility company with operations in more than 100 countries, where he was responsible for developing its broad international activities. Lernout, Hauspie and Willaert subsequently were arrested in Belgium, and Bastiaens was arrested outside of Boston and flown to Belgium on an outstanding arrest warrant. Seo reportedly has disappeared.

L&H filed for Chapter 11 bankruptcy in November 2000, and trading was suspended and then delisted on the Nasdaq, although it continues to trade on the Over the Counter "Pink Sheets" market. Under Chapter 11, L&H remains in control of its operations while it tries to figure out some type of payment plan with its creditors. In late May, L&H submitted a recovery plan to the Belgium court that was overwhelmingly approved by creditors but rejected by the court. Judge Michel Handschoewerker said the plan "lacks the necessary transparency toward creditors and it excels in vague and theoretical construction based on wishful thinking." The company's temporary protection from bankruptcy was extended until the end of September, and a subsequent plan has been submitted and was under review as this issue went to press. The original plan had called for the sale of its Mendez translation subsidiary, which L&H had been shopping for months with no buyer lined up. Mendez was finally sold to Bowne & Co. (New York) for $44 million, a far cry from the $160 million L&H was hoping for. The recovery plan also called for the sale of the Healthcare Solutions Group, formerly known as Dictaphone, and stressed that all assets were for sale if there was any interest. L&H also announced it was planning to lay off 1,200 employees to help curb costs, but said it was now a non-issue, as it already has lost 1,500 employees, a substantial number of whom were key employees who left on their own.

Other suppliers remain

With the implosion of Lernout & Hauspie and the interest in voice recognition systems apparently set to surge, the question becomes one of who is left in the market as a viable supplier? Is there anyone who would be interested in purchasing L&H's voice recognition technology? The answer to the second question is "hardly". L&H's competitors in the U.S. health care market include IBM, Nuance Communications (Menlo Park, California), SpeechWorks International (Boston, Massachusetts) and Phillips Speech Processing (Amsterdam, the Netherlands). Both IBM and Phillips are established companies with no apparent need of L&H's technology, or the desire to take on either its huge debt or legal problems. The other two VR companies have their own problems. Nuance is tangled in its own web of alleged investment fraud. Although not nearly as severe as L&H's, it nevertheless is the subject of a class-action suit. Nuance's stock price went from a high of $182 in August 2000 to under $10 a share in March 2001. It rebounded slightly, only to lose ground and currently is off over 90% of its high-water mark. SpeechWorks is in no shape to acquire the technology either, as it has seen its share price drop from a high of $108 to under $5. It is probably for this reason that Lernout decided to focus on selling Dictaphone and not its speech technology, as there really do not appear to be any suitable buyers.

Somewhere in here is a lesson on investing in foreign companies with big promises but little reporting, as well as the adequacy of big accounting firms' ability to uncover gross fraud. There is also the relevant question of what this means for medical VR applications The use of voice recognition in EMR systems ranges from documenting patient history, symptoms and problems to automated billing processes. So far there are a few systems that depend on voice recognition completely for electronic charting. By incorporating VR products, physicians have the flexibility of dictating free text while keeping transcription costs at a minimum. With the exception of some vendors such as Epic Systems (Madison, Wisconsin), which solely integrated Dragon's NaturallySpeaking software into its EpicCare charting system, most EMR companies have learned not to put all their VR eggs in one company's basket. Most computer-based record vendors on the market today can use any Windows-based voice recognition system, thereby factoring out the risks of integrating a bankrupt company's product, as well as passing the cost savings on to the customer. This approach means that customers do not have to live with a specific system, but can upgrade to better technologies as they become available, without revamping their whole electronic medical record.

Advanced systems available

Two companies that seem to be offering advanced systems include Nexvoice (Marietta, Georgia) and A-Life Medical (San Diego, California). Nexvoice's flagship software system, Vdocs, uses voice recognition to empower its EMR. Nexvoice's system can be integrated with any voice engine such as Viavoice, Dragon, etc. Vdocs uses speech macros to generate prescriptions, voice-enable clinical workflow, as well as command and control of any Windows-based PC applications. Not content to use voice as an add-on feature, Nexvoice claims Vdocs is the only true voice interaction EMR now on the market. Using "voice normals," macros that allow a user to input frequently used text into larger documents, Vdocs cuts the dictation process significantly as well as increases the accuracy of patient medical records. For those physicians needing greater mobility, Vdocs 2.0 also will support tablet computer hardware. As with VR technology in general, two limitations of this system are the accuracy of speech-to-text capabilities and the hardware cost to implement a system. VR systems depend upon fast processors and 100s of megabytes of memory. To accommodate multiple users on the system at the same time, the computer needs to be state-of-the-art.

A-Life's LifeCode technology, which was awarded "Best in Practical Innovations of Intelligent Systems Applications" from the American Association for Artificial Intelligence, uses natural language processing to interpret written text. Not only able to "read," but also "understand" dictated text, LifeCode extracts the dictated text and returns computerized ICD-9 coding results with speed and accuracy unknown in the coding industry. Although they are independent companies with independent applications, imagine what an EMR system like Nexvoice could accomplish if it incorporated a coding technology like LifeCode, which already has been incorporated into the new product offering from Siemens' SMS division (Malvern, Pennsylvania).

With growing security requirements mandated by HIPAA legislation and users of portable, wireless devices growing, voice recognition also is being touted as a cheaper and more mobile means of user authentication for computer-based systems. Voicevault by Buytel (Baltimore, Maryland) is one such system that is currently available. With a company-reported error rate of less than 1%, little to no capital outlay on hardware and speed of verification, it could be a useful and cost-effective tool. There is no mention, however, of how efficient this technology is at recognizing a recorded voice, which could make it no more secure than a system where cards are stolen and passwords are swiped all the time. However, Voicevault is cheaper than other authentication methods such as iris scanning and fingerprint reading, and is as portable as one's voice.

We are heading into an era where variations of voice recognition will permeate not only the medical community, but also modern society as a whole. However, Lernout & Hauspie seems unlikely to be there to greet us on the other side of the rainbow.