A Sinusoidal Noise Model Based Speech Synthesis For Phoneme Transition

One well-known problem with speech synthesis is the occurrence of audible discontinuities at phoneme boundaries, which lead to the unnaturalness of synthetic speech. This paper presents a sinusoidal noise based mathematical method to reform the transition regions from one phoneme to another phoneme with low storage. The speech parameters of sinusoidal noise model were estimated and stored as polynomials to reconstruct the transition wave. According to the results, all transitions regions which are considered during this experiment have higher correlation values for lower order polynomial with less capacity ratio. In addition, to that the same experiment has been carried out by changing the number of FFT coefficient. As the FFT coefficient increases, capacity ratio was also increased, while correlation coefficient values were also increased. It was understood that a signal which is very close to the original signal can be generated with a lesser number of FFT coefficients