Design Of Embedded Speech Processing System And Its Applicaton In Pitch Detection

Speech processing is one of the most important research fields, mainly including speech recognition, speech encoding or decoding, speech synthesis and other discipline branches. Along with the development of the embedded system and increasing understanding of that people to the value of speech interaction, embedded speech processing technology is gradually becoming a new research hotspot. As the core of embedded system, digital signal processor(DSP), which mainly faces the digital signal operation applications, is used more widely.Taking embedded speech processing for research background, by analyzing the function requirements of speech processing system, understanding the hardware structure and characteristic of C55x DSP, considering many factors, we put forward a overall system design measure, and select TMS320VC5502as the core to build a hardware platform of embedded speech processing system.The paper introduces the design principle and process in detail of power supply circuit, system decoding logic circuit, memory unit, reset and clock circuit, audio processing unit, communication interface unit, human-computer interaction unit, JTAG simulation unit and expand interface unit. In the memory design part, FLASH design, SDRAM design and EEPROM design are introduced. The communication interface unit design includes RS232interface design, RS485/RS422interface design and USB host-slave interface design. The signal integrity of the total circuit is also considered in the design process. We adjust the circuit to desire goal according to the analysing results. The paper expounds testing flow of the key module circuits in system with CCS development platform, and discusses preliminarily the system bootload based on FLASH.Pitch detection is a very important part of speech signal processing field, which has some application value in research of speech recognition or speech synthesis. Therefore, it’s necessary to transplant and realize the detection algorithm on the designed hardware platform. By comparing the common detection algorithm, we choose autocorrelation function method, and add some pre-processing and post-processing measures. After designing and optimizing the fixed-point C program, we realize transplantation and test of the algorithm on platform through alloting memory space, writing system initial program and establishing double buffers. The entire realization thoughts of pitch detection and double buffers structure based on DMA technology are common. So they can also be used for the real time realization research of other speech signal processing algorithm in the future.