Simple Technique for Speech Transposition in Real Time

If natural speech is filtered so as to remove all formant components except the second formant and then clipped, the resulting signal is highly intelligible. Transposition to lower frequencies is achieved by dividing this clipped signal by two using a simple digital divide-by-2 technique; spectrograms of the resulting signal for speech inputs reveal that the second formant band of the original speech has been transposed down in frequency by 1 oct. A modulator is used to reimpose either the original speech envelope or the envelope of the filtered speech onto the transposed signal. Finally, a low-pass filter having a cutoff of 1250 Hz and an attenuation slope of 48 dB/oct is used to prevent higher-frequency components from reaching the output. Scores on listening tests typically show a sudden increase from near zero to an acceptable figure after listener familiarization with the device output. [Work supported by NASA.]