NEC Corp and NEC Electronics Corp have developed software that enables mobile phone application processors to quickly respond to speech.
The test model is said to be able to recognize natural speech from conventional conversation, instead of recognizing each single word, using the built-in dictionary which contains tens of thousands of words.
The companies used the MP211 manufactured by NEC Electronics -- a processor featuring multiple CPU cores. The operating frequency is 200MHz. The new technology accelerates processing by dividing a course of voice recognition processing into phases, all of which are individually processed at the same time by the CPU cores on the MP211.
A course of processing is divided into the following three tasks: calculating the similarity between the input voice and the model sound reference; refining the search area; and determining an optimal word sequence from the refined search area. Overall recognition processing is reorganized into this three-step pipeline. Individual CPU cores perform each task simultaneously.
The companies have said that this technology realizes realtime voice applications for mobile phones without using any extra peripherals. They expect the voice recognition software to be used to look up the appropriate pages in e-manuals available through the mobile phone. The companies have made a prototype system that can immediately display the relevant pages in these e-manuals. If a user verbally requests, for example, information about how to change an email address, the relevant e-manual pages are displayed.
From NE Asia Online