Realization Of Speech Recongnition System On OMAP In The Framework Of Dsp Gateway

The application requirement in the field of electronic and mobile communication is experiencing unprecedented growth, but also putting forward higher requirements for embedded devices. Texas Instruments Open Multimedia Application Platform (OMAP) with asymmetric dual-core structure balances a variety of demand standards well. The processor combines the advantages of high performance, small size and low power consumption. OMAP with highly integrated hardware and software, is very suitable for complex tasks of the audio, video and other multimedia signals processing, due to the special optimization for information processing.Asymmetric dual-core structure as the greatest feature, the dual-core communication is the key technology of OMAP accordingly. The current solution is TI DSP/BIOS Link or Nokia DSP Gateway. The framework of DSP Gateway is more convenient and powerful relatively, but not widely used. The subject introduces this powerful tool and sets up embedded development platform on OMAP in the framework. Then the speech recognition system is implemented on the platform successfully, to validate and introduce the efficient embedded development platform architecture.The project researches the hardware and software characteristics of OMAP5912, and builds the hardware system with the OMAP5912development board as the core. The software development platform is set up around the framework of DSP Gateway. The project studies the textual configuration tool (Tconf) in detail, finding an easy way to re-generate the kernel library and to modify DSP configuration.With the deep analysis and research on the framework of DSP Gateway, the project proposes the application solution of the framework. The project solves the problems in application of endianism, DSP task structure design et cetera and successfully implements the large block data effective transfer between the dual-core, laying the solid foundation for the subsequent development and application.The neural network model which emulates biological principles has incomparable advantages in the intelligentize side, compared to traditional methods. As the basis of speech recognition system and task allocation between the dual-core, the theory and algorithm structure of speech recognition based on neural network are studied and analyzed in the project. The project describes the theoretical principle and design of program specifically, then realizes the speech recognition program on the ARM-side successfully.Based on the previous work, the speech recognition system in the framework of DSP Gateway is developed. The project designs three-stage program structure including speech feature extraction, network training and network recognition, then designs structures of dual-core program and DSP task for each part. Time-consuming convolution is optimized with assembly-optimized library DSPLIB. Eventually, the speech recognition system is implemented on OMAP in the framework of DSP Gateway. The correctness and timeliness are validated by results in comparison with those on ARM. The project introduces the framework of DSP Gateway to OMAP successfully.The OMAP embedded development platform in the framework of DSP Gateway built in the project is generally applicable for the audio, video and other multimedia signals, and other compute-intensive algorithms processing. The work of this project is significant in implementing speech recognition system based on OMAP embedded platform, and further more, provides an efficient and facile development environment to embedded system designers for reference.