Inverse analysis of vocal sound source using an analytical model of the vocal tract

Diseases occurring near the vocal cords, such as laryngeal cancer, share the initial symptom of hoarseness of voice. The GRBAS (grade, roughness, breathiness, asthenia, strain) scale is used as an acoustic diagnostic method for these diseases, but its objectivity is not well established. Instead, more accurate diagnosis may be possible by capturing the waveform of the volume velocity at the vocal cords. The aim of this study is to enable voice disturbances to be diagnosed by identifying the sound-source waveform from voice measurements. For acoustic analysis of the vocal tract, we modeled the air inside it as concentrated masses connected by linear springs and dampers. We identified the shape of the vocal tract by making the natural frequencies of the analytical model correspond to the measured formant frequencies, and we calculated the sound-source waveform from the measured voice waveform. To assess the validity of the model, we measured actual voices and used the model to identify the vocal tract shapes and corresponding sound-source waveforms. The identified waveforms have an asymmetrical triangular form, which is a feature of actual human sound-source waveforms. Local solutions allow multiple vocal tract shapes to be identified from a single sample. However, mathematical analysis showed that these differ only in the amplitude of the sound-source waveform, which does not affect the waveform shape. Furthermore, we built an experimental device that simulates the human voice mechanism and comprises an acrylic vocal tract and a piston. We confirmed that the identified sound sources are similar to measured sound sources. We therefore conclude that our proposed methods are valid.