Adaptive Blind Audio Signal Separation System using a TI DSP

Blind Signal Separation (BSS) deals with the problem of separating independent sources from their observed mixtures while both the mixing process and original sources are unknown. Examples of BSS algorithms employed in acoustical applications can be found among others in audio teleconferencing systems, hearing aids, noise cancelling etc. Such systems usually involve speakers that are at some unknown, time varying, distances from the microphones. Therefore the recorded microphone signals contain the unknown original sources mixed by unknown acoustical transfer functions. The main objective of such a system is to produce speech with a high-quality intelligibility despite a low signal-noise ratio of the observations. Above that, in contrast to many other techniques (e.g. beamforming) the blind approach is independent of the actual source locations. In this project such a high quality is achieved by using algorithms that operate across the entire audio spectrum. The complexity of the entire system is related to the room acoustics. The acoustic model of a room is very complex. In addition, due to the continuously changing nature of the acoustics, the (un)mixing system is time varying. Therefore, adaptive filters are required. Recently in our research group a new adaptive blind signal separation algorithm, which still is in research, has been introduced into literature [Yin]. This algorithm is based on a simplified mixing model and second order statistics. The simplified mixing model uses the property that the acoustic transfer functions from a source to two closely spaced microphones are very similar, therefore only a difference between these two functions is needed. The algorithm is efficiently realized in frequency domain. The main purpose of this project is to make a realization of this Adaptive Blind Audio Signal Separation algorithm on a Texas Instruments (TITM) TMD326006701 Evaluation Module (EVM) equipped with the TMS320C6701 floating point DSP, using Code Composer Studio as software development tool. With this realization the new algorithm can be developed further because realistic experiments can be done in real time. In order to do these experiments, also a flexible and user friendly Graphical User Interface (GUI) has been built. The GUI makes use of the Real Time Data eXchange (RTDX) protocol developed by TI. Experiments show that with A-BLASSTI two independent simultaneously occurring audio signals can be separated in real time. “This document was an entry in the TI DSP Challenge 2000, an annual contest organized by TI to encourage students from around the world to find innovative ways to use DSPs. For more information on the TI DSP Challenge 2000, see TI’s World Wide Web site at www.ti.com/sc/dsp_challenge.”