A Mechanical Voice System:Construction of Vocal Cords and its Pitch Control
Toshio Higashimoto and Hideyuki Sawada
Dept. of Intelligent Mechanical Systems Engineering, Faculty of Engineering, Kagawa University2217-20, Hayashi-cho, Takamatsu-city, Kagawa, 761-0396, JAPAN
A mechanical model of the human vocal system is being developed based on a mechatronicstechnology under a feedback control. While various ways of vocal sound production have been activelystudied so far, a mechanical construction of vocal system is considered to advantageously realize naturalvocalization with its fluid dynamics. Several motors are used to manipulate the mechanical voicesystem. The system is able to learn the relations between motor control papameters and the producedvocal sounds using an auditory feedback with neural networks, by mimicking a human vocalization.This paper introduces the construction of vocal cords and its adaptive control for the pitch learning. Themechanical system generates vowel and consonant sounds of different pitches by dynamicallycontrolling the vocal cords, vocal tract and nasal cavity.
Only humans are able to use words as theprimary media in the verbal communication, whilealmost all animals have voices. Different vocalsounds are generated by the complex movements of vocal organs under the feedback controlmechanisms using an auditory system. Vocalsounds and human vocalization mechanisms haveinterested many researchers for a long time
, andcomputerized voice production and recognitionhave become the essential technologies in therecent developments of flexible human-machineinterfaces studies.In the researches of sound production, variousways and techniques have been reported.Algorithmic syntheses have taken the place of analogue circuit syntheses and became widely usedtechniques
. Sound sampling methods andphysical model based syntheses are typicaltechniques, which are expected to provide differenttypes of realistic vocal sounds
. In addition tothese algorithmic synthesis techniques, amechanical approach using a phonetic or vocalmodel imitating the human vocalization mechanismwould be a valuable and notable objective.Several mechanical constructions of a humanvocal system to realize human-like speech havebeen reported
. In most of the researches,however, the mechanical reproductions of thehuman vocal system were mainly directed byreferring to X-ray images and FEM analysis, andcontrol methods for natural vocalization have notbeen considered so far. In fact, since the behaviorsof vocal organs have not been sufficientlyinvestigated due to the nonlinear factors of fluiddynamics yet to be overcome, the control of mechanical system has often the difficulties to beestablished.We are developing a mechanical voicegeneration system together with its controlmechanism for voice production imitating humanvocalization. The fundamental frequency and thespectrum envelope determine the principalcharacteristics of a sound. The former is thecharacteristic of a source sound generated by avibrating object, and the latter is operated by thework of the resonance effects. In vocalization, thevibration of vocal cords generates a source sound,and then the sound wave is led to a vocal tract,which works as a filter to determine the spectrumenvelope.We have constructed a motor-controlledmechanical model with a vocal cord and a vocaltract so far
. By introducing an auditoryfeedback mechanism with an adaptive controlalgorithm of pitch and phoneme, the system is ableto autonomously acquire the control method of themechanical system to produce stable vocal soundsimitating human vocalization
. In the system, anartificial vocal cord used by people who had toremove their vocal cords because of a glottaldisease was used. The vibration of a rubber with5mm width stretched over a plastic body madevocal sound source. The tension of the rubber wasmanipulated by applying tensile force with a motor,so that the fundamental frequency of a generatedvocal sound was changed easily.We paid attention to the quality of a soundgenerated by the voice system to be close to ahuman, and worked to develop human-like vocalcords. This paper describes the construction of vocal cords and its control for changingfundamental frequency, together with an adaptivelearning mechanism.