Vocoded Speech in the Absence of the Laryngeal Frequency

Most pitch excited channel vocoders require the fundamental or laryngeal frequency of the input speech to be present if the output speech is to be of high quality. In order to determine if speech whose fundamental is absent can have its pitch accurately restored so as to be used as an input to a vocoder, a computer simulation was performed. The fundamental was restored by passing the speech through a fullwave rectifier followed by a slope filter. The accuracy of the pitch restoration of this method was compared with that of simply measuring the pitch of speech whose fundamental was present by slope filtering alone. A third pitch detection method, that of visually displaying the speech waveform and determining the pitch by eye, was also used as a comparison. Pitch contours of the three methods indicate that pitch restored by fullwave rectification and slope filtering has larger perturbations than pitch as detected by slope filtering alone. Both methods produced pitch contours having much larger perturbations than pitch determined visually. Speech whose pitch was determined by the above three methods was used to excite the spectrally flattened Lincoln Laboratory Vocoder. Listening tests of the vocoder output indicate that pitch restored by fullwave rectification and slope filtering produced rougher sounding speech than speech whose pitch was detected by slope filtering alone, but both methods produced speech having considerably more audible roughness than that produced by visually detected pitch. Finally, the sophisticated pitch detector of the vocoder itself produced speech of quality comparable to that determined visually. Accepted for the Air Force Franklin C. Hudson Chief, Lincoln Laboratory Office