Energy-driven distribution of signal processing applications across wireless sensor networks
ACM Transactions on Sensor Networks (TOSN)
| Hi-index | 0.00 |
We present a new design of an Embedded Speech RecognitionSystem. It combines the aspects of both hardwareand software design to implement a speaker dependent, isolatedword, small vocabulary speech recognition system.The feature extraction is based on modified Mel-scaled FrequencyCepstral Coefficients (MFCC) and template matchingemploys Dynamic Time Warping (DTW). A novel algorithmhas been used to improve the detection of start of aword. The hardware is built around the industry standardTMS320LF2407A DSP. The board is designed to serve asa general purpose DSP development board for the 24X seriesof TI DSPs. It contains, apart from the DSP, the externalSRAM, FLASH, ADC interface, I/O interfacing blocksand JTAG interface. Both the hardware and the softwarehave been designed concurrently, with a view to achievehigh-speed recognition with maximum accuracy in minimumpower and making the device portable. The proposedsolution is a low-cost, high-performance, scalable alternativeto other existing products.