Porting A Client Agent For Pitch Estimation and Noise Suppression To An Embedded Multi Processor

This paper is to implement contents based music retrieval method on the embedded digital signal processor (DSP) chipset. This method is basically Query by Humming and Singing (QbSH) with polyphonic music signal. The client has to extract the feature and suppress the noise from user humming signal recorded with input device. Noise suppression algorithm makes merge of Minimum Statistics (MS) for stationary noise and Improved Minima Controlled Recursive Averaging (IMCRA) for non￢stationary noise, and the feature is estimated with temporal and spectral autocorrelation simultaneously to reduce the pitch having and doubling problem. We develop the prototypes of clients for three different platforms like computer, smart phone/pad and KARAOKE. The client part of this system is ported on the DSP chipset which is the TI OMAP processor named L138 including C674x core up to 300㎒ clock cycle for DSP part and ARM-9 core up to 600㎒ clock cycle.