Human Machine Speech Interface System Based On OMAP5912

Absolutely current mobile electronic devices are becoming smaller and more complicated. So it is inconvenient for users to play with these devices whether by keyboard or touch screen, especially for an aged or blind person. Beside that, the similar thing also happens in the manufacture, in the army, as well as in the hospital and many other places. Speech is the easiest and most convenient way for human beings to communicate, so those problems shall be well solved if a Human Machine Speech Interface (HMSI) system exists inside these devices. Embedded Speech Recognition System (ESRS) plays a key role in HMSI system and has a large market. Because of the computing resource shortage and unpredictable environment character of an embedded system, ESRS is still an international big problem of speech recognition technology. This paper mainly focused on speech recognition problems on embedded devices, and succeeded in building a HMSI system on OMAP5912 platform.When making the hardware decision, the embedded processor should support GUI and some other programs when HMSI system was running, it should have lower power consumption at the same time. Then the original single core processor can hardly meet all of those requirements. We took the asymmetrical dual core OMAP5912 of TI as our embedded CPU when its TMS320C55x DSP core is very suitable for digital signal processing because of its ultra low power consumption. So the DSP can share some computing task from GPP, improve the HMSI ability and the whole system shouldn't get a lot watt-loss at the same time.When making the software decision, we found that open source and GNU software has been widely used by scientists for over 20 years. So the system was based on available open source software to largely reduce development cycle as well as money costs. We built a software development platform for OMAP5912, ported U-boot, Linux kernel and DSP Gateway to OMAP5912, demonstrated that DSP runs well and GPP could control its running by DSP Gateway, so DSP can share some signal processing tasks from GPP during the speech recognition progress.We chose PocketSphinx to be the speech recognition engine for the HMSI system. Some too complicated parts of the source code were removed while some functions were merged, modified or enhanced. These acts would simplify the running or keeping of the code. Finally we ported PocketSphinx speech recognition engine to OMAP5912 processor, realized a HMSI system on it, and accomplished our plan.