CLASSIFICATION OF SPECTROGRAPHIC VOICE PA TTERNS USING SELF-ORGANIZING NEURONAL NETWORKS (KOHONEN MAPS) IN THE EVALUATION OF THE INFANT CRYWITH AND WITHOUT TIME-DELAYED FEEDBACK

Subjective and auditory assessment of the voice is now more commonly b eing replaced by objective voice analysis. Because of the amount of da ta available from computer-aided voice analysis, subjective selection and interpretation of single data sets remain a matter of experience o f the individual investigator. Since neuronal networks an widely used in telecommunication and speech recognition, we applied self-organizin g Kohonen networks to classify voice patterns. In the phase of ''learn ing,'' the Kohonen map is adapted to patterns of the primary signals o btained. If, in the phase of using the map, the input signal hits the field of the primary signals, it will resemble them closely. In this s tudy, we recorded newborn and young infant, cries using a DAT recorder and a high-quality microphone. The cries were elicted by wearing unco mfortable headphones (''cries of discomfort''). Spectrographic charact eristics of the cries were classified by 20-step bark spectra and then applied to the neuronal networks. It was possible to recognize simila rities of different cries of the same children and interindividual dif ferences, as well as cries of children with profound hearing loss. In addition, delayed auditory feedback at 80 dB SL was presented to 27 ch ildren via headphone using a three-headed tape-recorder as a model for induced individual cry changes. However, it was not possible to class ify short-term changes as in a delayed feeback procedure. Nevertheless , neuronal networks may be helpful as an additional tool in spectrogra phic voice analysis.