摘要 |
PURPOSE: To surely and automatically detect a phrase boundary by computing an evaluation function value for every plural phrase boundary and making the phrase boundary combination, at which each evaluation function value becomes a minimum, a detection result. CONSTITUTION: The device is provided with an acoustic data selecting section 1 which selects the acoustic feature parameter data based on the inputted character strings by referring to the voice data-base in a voice data-base memory 12, a parameter group producing section 2 which produces a feature parameter group for voice synthesis and a voice synthesizing section 3 which produces audio signals based on the produced feature parameter group and outputs the signals to a speaker 4. Then, prescribed weighting coefficients are respectively multiplied to the AIC(Akaike Information Criterion) reference and the estimated errors of pitch frequencies and the result is added to obtain an evaluation function value. The value is computed for every plural phrase boundary which is to be detected in one sentence and a detection result is obtained from the combination of the phrase boundary at which each evaluation function value becomes a minimum. |