New Applications of Voice-Excitation to Vocoders

Abstract

The principle of voice-excitation for a vocoder synthesizer involves generation of a flat spectrum of pitch harmonics or noise from an uncoded subband (“baseband”) of the original speech [M. R. Schroeder and E. E. David, Jr., “A vocoder for transmitting 10 kc/sec speech over a 3.5 kc/sec channel,” Acustica 10, No. 1, 35–43 (1960)]. The necessity for a decision-making pitch detector is thereby removed. The principal advantage of the voice-excited vocoder is its insensitivity to input signal-to-noise ratio and equalization. A voice-excited vocoder with a 720-cps (250–970 cps) baseband and 17 spectrum channels low-passed to 25 cps each, covering the band 970–3700 cps, has been built and evaluated. The test shows an average PB-word intelligibility of 86%, compared to 92% for input speech of the same bandwidth, both with and 18-db signal-to-noise ratio. Quality tests indicate that about 85% of all listeners rate the speech “as good as” the input while only 20% so consider a vocoder using conventional pitch control. The vocoder performed about equally well for each of the 12 speakers in the quality test. Voice-discrimination tests indicate that voice identity is well preserved. Crucial factors influencing the remade speech quality are the accuracy of spectral flattening and the impulse response of the analyzer low-pass filters. Studies of voice-excitation indicate that the baseband, if used for excitation only, can be coded into about 2000 binary pulses per second. Accordingly, the vocoder was modified with the addition of four spectrum channels, and operated with a digitally coded baseband. In total, it is estimated that about 4000 binary pulses per second are required for all-digital transmission. A tape demonstrating these results will be played.