Free Speech: Philips Joins Continuous Dictation Race

Philips has introduced a large vocabulary, general English dictation system for business, correspondence, and general use. FreeSpeech98 is the fourth new dictation system to come to market along with IBM, Dragon, and Lernout & Hauspie. Philips was actually the first to offer commercially available continuous speech recognition in the vertical market for radiology and other applications. FreeSpeech98 is a full featured, general dictation system designed for business and general use. Philips rightly encourages an “enrollment” period in which one trains the machine for an hour or more in order for it to learn the speaker’s voice pattern. The system requirements for Philip’s FreeSpeech98 for the Windows 95 and 98 versions are a Pentium MMX 166 MHz 32 MB RAM; recommended is a Pentium at MMX 200 MHz 64 MB RAM. And for the Windows NT 4.0, minimum requirements are a Pentium MMX 166 MHz, 48 MB RAM; recommended hardware is a Pentium MMX 200 MHz, 96 MB RAM. Only one speaker is permitted per copy. The software is free for a week, and then requires a fee of $39. We tested FreeSpeech98 on a Pentium 200 megahertz MMX with 48 megabytes of RAM. After the initial training period, the accuracy was greater than the L&H system, but not as good as the IBM or Dragon systems. On launching the program, which opens moderately quickly, a command bar appears along the top or side of the screen. Then the user starts an application such as WordPad, Microsoft Word, email program or most Windows software programs. While we were dictating into WordPad, the system crashed resulting in the total loss of our document. Periodic backups during dictation might help solve this problem. Working from a separate recording also allows one to keep a backup record. We asked the Philips system to recognize a wave file recording and it did so with admirable speed transcribing the recorded material. Unlike the other systems, Philips needs to be instructed to use particular corrected documents for improvement of one’s speech profile. After correcting a dictated document, one has the option of allowing the Philips product to incorporate these pronunciation corrections into the user’s speech model. Moreover one is supposed to avoid formatting the text prior to designating it for speech model updating. With Philips, the updating of one’s speech model actually becomes a multi-step task requiring using several different series of buttons. This difficult, multi-step process can prevent uncorrected documents from possibly corrupting or worsening one’s speech files, resulting in a poorer recognition rate. However it could also discourage users from updating their speech files, which is done automatically by the other systems. Philips has a nice method of playing back recorded speech behind typed words. Highlighting bounces from word to word indicating which word correlates with which sound, easing the correction process. One can give the playback command by voice, by a mouse click, or by pressing a button on their special microphone, the Speech Mike Pro. The Speech Mike Pro is a high-quality microphone designed for dictation with a speaker unit, and a trackball for mouse movement. A simpler version of this microphone is also available, called simply the Speech Mike. This current version of FreeSpeech98 only allows one voice profile for one user. No doubt future editions will allow multiple users as well as multiple profiles for each user. In our hands, this Philips system appeared to have a significantly lower accuracy rate than Dragon or IBM, requiring a frustratingly larger number of corrections. The accuracy rate was lower than IBM and Dragon after initial training of those products, but higher than Lernout & Hauspie’s system. The accuracy rate after continued use remains to be determined, and is most important. The recommended price of this software is $39 but does not include a microphone. Using a high-quality microphone designed for speech recognition dictation is highly recommended for use in almost any environment. Vertical market versions of this product for various vocabularies in medicine and law and other areas are available through partners of Philips, including networked versions. Microphones
A distinguishing feature of the Philips system is the integration with the Speech Mike Pro, an accessory which is optional and must be ordered separately. This microphone/speaker device allows one to dictate, playback, and correct while using a single handheld unit as opposed to a headset unit. It also contains a mouse trackball, as well as special buttons to begin and end dictation, and to move the cursor forwards and backwards, word by word, and to initiate and stop playback of the recorded voice. There is also a plain Speech Mike available which includes the microphone speaker and trackball: a handheld device without the dictation playback and correction buttons. The Speech Mike is also an excellent sensitive dictation microphone which can be used admirably with other speech recognition systems. It could also be used for Internet telephony. The Speech Mike and Speech Mike Pro are very exciting microphones for general speech recognition use.

---

Peter Fleming, a speech recognition consultant, may be reached at aris@world.std.com, or (617) 923-9356.