Embedded system for audio source separation

The paper presents an embedded system, designed for audio signal processing. It consists of a microphone array, the signal conditioning circuits, the AD converter, 2 DSP modules and the communication circuits, placed on a single board (excepting the microphone array). Up to 8 microphones receive the audio signals, which are recorded and processed, in order to provide information about the audio sources in the area. The processing modules make use of 2 powerful DSPs of the family Blackfin (Analog Devices), with up to 600 MHz, 64 MB RAM and 8 MB flash. The board communicates with other boards via a native Ethernet interface of one of the DSPs. This paper considers the board included in a source separation application. A beamforming algorithm is used, in order to select the audio sources laying in a specific direction. Other appropriate tasks for this board are: source localization, blind source separation, audio events recognition. Usual application areas are surveillance, security and noise reduction. The properties of the system are experimentally evaluated. It satisfies the requirements of such an application and it is very reliable.