Helium speech enhancement by frequency-domain processing

The breathing gas of divers is a pressurized mixture of helium and oxygen that distorts the divers´ speech. The formant frequencies and bandwidths shift upward in a nonlinear manner while the pitch period remains relatively unchanged. Commercial unscramblers (voice enhancers) work in the time domain, sampling segments of diver speech and expanding them. This technique does enhance diver speech, but not enough for reliable communication. We have implemented an algorithm that operates in the frequency domain and runs in real time on an Analogic AP400 array processor. The algorithm transforms helium speech segments into the frequency domain, compresses the formants back to normal speech values, and then transforms the corrected speech segments back into the time domain. Identical speech material was processed by both the time- and frequency-domain unscramblers and the results are compared.