Intel Previews Pentium III
Intel Corp. (Santa Clara, CA) assembled more than 250 exhibitors and as many members of the media in San Jose recently to preview the release of its next-generation, Pentium III chip. The new chip, code-named Katmai, is designed specifically to support the transmission of advanced multimedia technology over the Internet. It contains 70 new instructions, new architectures for floating-point and streaming operations, a unique ID code, and an enhanced cache layout that supports faster processing of large quantities of data. All of this runs at 450 to 550MHz. The preview began with multimedia (of course) presentations by Craig Barrett, Intel's president and CEO, and Mike Aymer, VP and Director of the Platform Launch Operations. This was followed by a visit to the exhibit hall. The meeting concluded with a question-and-answer session that showed the media are far more focused on privacy issues associated with the unique ID codes than they are with multimedia. Intel supplies both the codes and software to turn transmission of ID codes off or on. Based upon their answers to some of the questions about the ID codes, Intel's position appears to be that they have done all they could and, therefore, responsibility for safeguarding anonymity/privacy rests upon the individual or business entity. Given the ways in which cookies are now used, this argument was greeted with skepticism. In their introductory presentations, Barrett and Aymer illustrated the power of the Pentium III by showing stunningly beautiful, interactive demos that have already been deployed on the Internet by companies like The Sharper Image. There were no speech-processing demos, but Aymer concluded his enumeration of the functionality in the new chip by referring to its special support for speech recognition. Then, a giant blue door bathed in artificial smoke opened slowly to reveal the pathway to the exhibits. The exhibit hall was divided into ten islands organized according to themes, including web-enabled office/productivity, e-business, and creativity. Unfortunately, this organization made it difficult to easily locate companies using speech processing. Philips Speech, IBM, Lernout & Hauspie, and Lotus Notes with ViaVoice were in the Productivity island along with Dell and NEC Computer Systems while Dragon shared a booth with Compaq in the Development Environment island. Each speech-recognition company developed its own demos for the preview and those demos were especially interesting because they revealed ways that the features of the Pentium III could be exploited by speech recognition. Recognizing the need to use speech in conjunction with multiple applications, Philips Speech Systems developed a demo showing how its FreeSpeech technology would operate quickly and accurately for both dictation and command-and-control - even when several business-oriented application programs are loaded. The demo ran with five open applications, including Microsoft Word. Philips plans to release a new suite of speech products optimized for the Pentium III in mid-1999. IBM's demo system included voice-activated, intelligent-agent technology that has been implemented in the company's just-released ViaVoice SDK. The intelligent agent is always active. It can remain in the background or it can be shifted into the foreground, appearing as an animated figure that can respond to free-form voice commands. According to IBM, Pentium III power is not required for the voice-activated agent technology. However, because the chip reduces the computational cycles needed for the speech, it is possible to reduce complexity for the agent. Lernout & Hauspie demonstrated dictation followed by text-to-text translation and text-to-speech output of the translated text. The demo was run between an English-speaking participant at one PC and a German-speaking participant at another PC. According to Paul McNulty, VP and General Manager of L&H's PC Applications Group "We firmly believe that the user of our translation software will benefit from the increased processing speed of the Pentium III chip." He added, however, that the company has not yet completed its evaluation of the new instruction set for translation. The Dragon demo highlighted the speed with which dictation technology can be applied to Internet navigation. The company has also applied Pentium III power to reducing enrollment time for the user. Dragon has reconfigured its models and algorithms to exploit the Pentium III speed and better support of floating point operations to cut the average user recording time for enrollment from 18 minutes to 2 minutes. These changes enable the software to extract and process more information from each slice of speech. Evaluation
As a group, the speech companies I spoke with were favorable but reserved in their evaluation of the Pentium III. All of them reported that it offers tremendous improvements in processing speed. Some also mentioned the chip's improved support for floating point operations as a valuable feature and one company, Lernout & Hauspie, highlighted the enhanced caching for rapidly anticipating and utilizing large amounts of data. They admitted that the increase in speed has made it possible to perform more sophisticated operations, but they had not had the time to assess the full import of the functionality changes for speech recognition. Furthermore, most of them reported they had not yet begun to determine the benefits - if any -- the Pentium III might provide to text-to-speech synthesis, translation, speaker verification, or natural language processing. It will take time and work to determine the extent to which the Pentium III benefits speech and language processing. If, as Intel suggests, the chip represents a qualitative difference, then the final assessment must wait until the speech companies change their technology to exploit that difference.
Judith Markowitz is president of J. Markowitz Consultants, and can be reached at Northwestern University/Evanston Research Park, 1840 North Oak, Evanston, Ill., 60201, or by e-mail at email@example.com. Her column Voice IDeas appears regularly in Speech Technology magazine.
Companies and Suppliers Mentioned