In the days when people used typewriters, typists dreamed of creating an automatic typewriter that would type anything on verbal command. For more than 35 years, engineers and speech specialists have been working to make that dream a reality in the form of a computer that recognizes speech. Although neither that dream nor the fantasy communication system from Star Trek has been achieved today, automatic speech recognition is used in toys and games, instruments and computers, data collection, and dictation.

Until recently, speech recognition as a technology was still imperfect and expensive, which led to the use of different recognition methods for different products or purposes. A computer chip that recognizes 10-15 words costs a few dollars, but the powerful computer and accompanying software required by early dictation systems costs thousands of dollars and recognizes only 50,000 words. With the introduction of Breveon™ AuratorPC, however, price points for dictation systems are dropping, while the quality of speech-recognition technology is increasing. Speech-recognition dictation software now costs a few hundred dollars and runs on computers costing less than $1,000. Some applications, such as an over-the-telephone information system, must recognize anyone's voice, but other applications, such as personal telephone dialing or dictation systems, are usually adapted by each user to recognize their voice and pronunciation.

Although many modern speech-recognition programs and devices voice-enable their systems, the terms voice recognition and speech recognition are not synonymous. While both use technology to capture the spoken word, voice recognition and speech recognition have different goals, and run different technologies. Speech recognition is continuous, natural language processing. In contrast, voice recognition uses recordings to determine an individual's identity, an Orwellian twist on today's social security number and fingerprint.

Speech-recognition technology works through a series of complicated algorithms that translate math into words and a program that processes the words. Smaller speech-recognition systems that recognize around 100 words compare sets of acoustic features that are measured in a user's speech with those stored in patterns in templates. But to make a system that will recognize whole vocabularies and multiple voices, developers of those systems must collect and process representative speech samples from many people. The samples then help developers create the phonetic engine, vocabulary, and language model, which are the three basic parts of speech-recognition technology.

Despite its name, a phonetic engine is a type of software that is programmed to recognize the smallest bits of sounds - called phonemes - uttered in American English. The phonetic engine then assembles these recognized phonemes into spoken words and translates them into written or computer-processed text.

The key to success in speech-recognition systems lies with the second component of the technology, or the language model and vocabulary. A language model is the way in which words are naturally used together, according to the grammar and spelling rules of that language.

Until recently, the production of reliable, efficient speech-recognition programs for medical dictation had been slow in coming and expensive upon arrival. Standard speech-recognition systems are unwieldy for the professional or specialized user, causing users to dictate for many hours with poor recognition accuracy until the software learns their vocabulary, and spend hours collecting and processing a great deal of representative, computer-readable text.

As an alternative to this time-consuming and costly method, Breveon (formerly KorTeam) has focused its efforts toward developing some of the most efficient and cost-effective vocabulary and language models on the market, so that professional and medical users can boost productivity right out of the box.

Because when all is said and done, you want your speech recognition program to work.