Keep these points in mind to achieve the best possible speech intelligibility:
The spectra of the analysis and synthesis signals should almost completely overlap. Coupling low male voices with synthesis signals in the treble range doesn’t work well.
The synthesis signal must be constantly sustained, without breaks. The incoming side-chain signal should be played or sung legato, because breaks in the synthesis signal stop the vocoder output. Alternatively, the Release parameter of the synthesis signal—not the Release time of the Analysis section—can be set to a longer time. You can also achieve nice effects by using a reverberation signal as a synthesis signal. Note that the two latter methods can lead to harmonic overlaps.
Do not overdrive the vocoder. This can happen easily, and distortion can occur.
Enunciate your speech clearly if the recording is to be used as an analysis signal. Spoken words with a relatively low pitch work better than sung vocals—even if the creation of vocoder choirs is your goal. Pronounce consonants well, as exemplified in the rolled “R” of “We are the Robots,” by Kraftwerk, a classic vocoder track. This exaggerated pronunciation was specifically made to cater to the vocoder.
You can freely set Formant parameters. Shifting, stretching, or compressing the formants has a minimal effect on the intelligibility of speech, as does the number of frequency bands. The reason for this is due to the human ability to differentiate the voices of children, women, and men, whose skulls and throats vary. Such physical differences cause variations in the formants that make up their voices. Human perception, or recognition, of speech is based on an analysis of the relationships between these formants. In the EVOC 20 plug-ins, these relationships are maintained even when extreme formant settings are used.