Speech recognition programs pin hopes on transcription

Voice recognition software companies are buying medical transcription firms in hopes of improving both industries.

By Tyler Chin, amednews staff. June 26, 2000.

Two major speech recognition software companies have acquired control of medical transcription companies in transactions that their buyers hope could improve the turnaround time for transcribed reports for physicians, as well as spur them to transmit records over the Internet.

The deals also are a reflection that speech recognition software, in which electronic medical records are filled out by the sound of physicians' voices, not the typing of their fingers, is not yet ready for prime time.

In the largest deal, Amsterdam, Netherlands-based Royal Philips Electronics announced on May 22 that it is acquiring 60% of MedQuist Inc., Marlton, N.J., for $1.2 billion.

On June 2, Lernout & Hauspie Speech Products N.V., Ieper, Belgium, agreed to pay about $25 million to acquire Broomfield, Colo.-based Rodeer Systems Inc.'s medical transcription operations in Arizona, Georgia, Minnesota, Oklahoma and Texas. The acquisition enables L&H to expand its existing medical transcription assets, which include Dictaphone Corp., Stratford, Conn.

By acquiring majority ownership of MedQuist -- one of the largest medical transcription companies in the United States -- Philips gets the services of MedQuist's 8,500 medical transcriptionists while MedQuist gets a partner that will help it realize its plan to upgrade its technology to an application service provider model from a DOS-based system, said John Donohoe, MedQuist's president and COO. DOS is an old operating system from Microsoft Corp. that was the precursor to Windows.

In the ASP model, which Lernout & Hauspie also is pursuing, a company makes its software available over the Internet typically for a monthly subscription fee.

The deals will enable MedQuist and L&H to grow their business and expand into the medical document services area, including data analysis, executives said.

Speech recognition not 100% accurate

The transactions, however, also reflect a recognition by medical transcription and speech recognition companies that they need each other to expand their business as transcription shifts from an analog to a digital platform in which digitized voice files are transmitted over the Internet.

"It is an acknowledgment [by Royal Philips and L&H] that although there have been many advances made in speech recognition, the technology is not ready to go live," said Adam Frisch, director at UBS Warburg, New York.

For years, continuous speech recognition companies, including Philips Speech Processing, a unit of Royal Philips, and L&H, have promoted the technology, in which speech is converted into text on a computer screen as a physician dictates a report into a microphone attached to the computer.

However, most doctors have not adopted the technology because it is not 100% accurate and it disrupts their work flow. The doctors who do use the technology typically work at academic medical centers, which tend to be early adopters of technology and can afford to pay for it.

Companies claim the accuracy rate of continuous speech recognition technology is as high as 95%, but that hasn't been good enough to prompt physicians to buy and use speech recognition systems in-house as a stand-alone product.

"Even if speech recognition is 90% accurate, ... that means 10% of the document is incorrect. Say there are 400 words on one page -- that's 40 mistakes per page!" said Donohoe, of MedQuist. Another problem is that the systems may spell words correctly but out of context, he said.

Despite these shortcomings, MedQuist and L&H said they are deploying speech technology on the back end of the transcription process because it will help boost efficiencies and deliver better service to physicians and other customers.

Boom in medical transcription market?

By blending speech recognition and traditional transcription services, the companies will be able to increase employee productivity and also overcome a shortage of qualified medical transcriptionists, the companies said.

They will be able to hire workers who won't have to transcribe reports from scratch, but instead will edit the already-transcribed reports to ensure accuracy, the companies said.

The new systems will substantially reduce the time it takes physicians to get transcribed reports, said Gaston Bastiaens, L&H's CEO and president. The turnaround time for traditional systems is 24 to 48 hours.

The transition to the Internet and speech-recognition technologies also will lower telecommunication costs, possibly resulting in lower transcription costs for physicians, who now pay 13 cents to 15 cents a line, Donohoe added.

The companies, of course, aren't embarking on these efforts to be nice to doctors: They see a huge business opportunity in the market for transcription service, which MedQuist estimates may exceed $15 billion.

MedQuist and L&H, which respectively reported $330 million and $344 million in revenue in 1999, expect to start rolling out new products and services by the end of the year.