ISCA Archive ECST 1987
ISCA Archive ECST 1987

Multisensor input for a speech enhancement equipment: system architecture and experimental results

Livio Busnelli, Neviano Dal Degan, Luigi Vetrano

Speech enhancement techniques based on two inputs algorithms are completely inadequate if ambient noise reveals strong incoherence. On the other hand, enhancement obtained by using spectral subtraction methods presents several inconveniences while also being rather complex. The herein presented equipment and the relevant algorithms use an array of input sensors to pick-up the speech in a noisy environment; appropriate signal processing permits to exploit both the spatial coherence of the speech and incoherence of noise. The signal to noise ratio of the output signal shows a very good accordance with that of the theoretical model. The processing load can be undertaken by currently available DSP technology.