BROADBAND COMMUNITIES is the leading source of information on digital and broadband technologies for buildings and communities. Our editorial aims to accelerate the deployment of Fiber-To-The-Home and Fiber-To-The-Premises.
Issue link: http://bbcmag.epubxp.com/i/622360
6 | BROADBAND COMMUNITIES | www.broadbandcommunities.com | NOVEMBER/DECEMBER 2015 BANDWIDTH HAWK T here's a huge amount of audio trafc coming to the Internet. Only fber provides the reliability and upward-bound bandwidth to really change how people interact with their computing appliances – or any appliances, for that matter. Few network operators have been paying attention, and it is easy to see why. When prognosticators look ahead, they usually confne their forecasts to "more of the same." In video, 4K is beginning to supplant HD, and people reasonably expect that 8K is lurking just below the horizon. Cloud storage of home video has grown by leaps and bounds, and everyone reasonably expects it to continue growing. Voice trafc hasn't yet become a big deal, and people reasonably expect that it never will. Tey're wrong. Users marvel that Siri, OK Google and Cortana can decipher almost any accent in any background-noise situation, but they may not be aware that the voice-to-text translation isn't carried out on their phones, tablets or desktop machines. Desktop computers have the power to slice analog speech into ever-fner digital segments, and that helps. But no machine at the user end has near-instantaneous access to a world of context (such as queries by others in the same location or situation) that helps make the translation to text startlingly almost perfect. As fewer people want or need to maintain desktop machines, the future scaling of this technology will occur on mobile devices, usually connected to in-building Wi-Fi, not to cellular networks. Tere are two ways to do that. Te most obvious is to increase computing power in handheld devices. Tat, in turn, requires more compact, reliable mobile power sources. One candidate, the fuel cell, looks good for a potential 10-fold increase in available mobile watt-hours for small devices such as phones. More efcient CPUs and even voice processors analogous to the graphics processors gamers love can double or triple that. However, truly taking advantage of voice processing requires worldwide, universal connectivity. Te fctional Starship Enterprise on the Star Trek TV and flm series had a great voice recognition and control system. How? It had a big enough on-board computer to store and instantaneously access the United Federation of Planets' combined knowledge base. Costly, but a military-size budget can do wonders! Earthbound users will want super-Siri and her friends to • Respond like Mom to their every need. • Provide typed transcripts of any spoken material – most people can read three times faster than they can listen. • Prepare multiple instances of spoken material for logical comparison. • Fully index the dialog in any video or collection of videos so it can be searched as easily as text. • Respond accurately to queries about an infnite number of domain spaces. • Handle nonlinear requests, bouncing from topic to topic during morning drive time. • Adjust anything in an increasingly adjustable world – lighting, temperature, humidity, water temperature – without knobs or touch screens. For more than a decade, trouble-free voice-to-text processing has been elusive. Te software has gotten much better, but for most users, the error rate is unacceptable for long documents. Tus, Google Keep and similar free voice- to-text apps are distributed for recording quick and somewhat error-prone notes on tasks or quick ideas, not for formal documents in which spelling and punctuation matters. Any one person dictating into a phone or tablet isn't using much bandwidth. But if even one in 1,000 people dictate at the same time, projected upstream bandwidth demand would at least triple, adding about 30 kilobytes per hour per user on average. By comparison, the IEEE projects that, by 2025, the much-discussed Internet of Tings will add just 7 kilobytes per hour per user to Wi-Fi demand. Users of smartphones (there are 1.5 billion in use worldwide today, with 3 billion expected by 2019) generate about half the worldwide mobile voice trafc, or 50 PB/month. Applications based on audio recognition may generate at least as much trafc and probably more. Siri alone can now book restaurant reservations and soon should be able to handle travel reservations. Tese and other useful audio applications will soon be so commonplace that people will wonder how they ever did without them. Wall Street will once again see a new set of uses stimulating broadband demand. v Contact the Hawk at email@example.com. The Next Bandwidth Hog Move over, video. Only fber to the home can handle the upstream trafc demands of emerging voice recognition services. Wall Street take note. By Steven S. Ross / Broadband Communities