赤木正人
音声知覚におけるスペクトルターゲット予測モデルの評価
Abstract:A model of a phoneme target prediction mechanism based on psychological and physiological knowledge is proposed and evaluated by comparing predicted values with results of auditory psychological experiments. When the trajectories of physical characteristics of speech in spectral space are approximated by a 2nd-order critical damping system, the proposed model can estimate target values using a short period speech wave (50ms) without knowing the onset positions of the spectral transition.
Additionally, this model decreases the length of transitional sounds which produce spurious vowels, and recovers vowel characteristics neutralized by co-articulation.
Moreover, the model compensates for the transitional part of syllables and achieves the extraction of stable acoustic features for consonants. This model is considered to be a good approximation of the target prediction mechanism in speech perception.