7th International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2011)

Florence, Italy
August 25-27, 2011

Estimation of Multiple Source Component using Genetic Algorithm

Cheolwoo Jo, Jaehee Kim

School of Mechatronics, Changwon National University, Korea

Source of speech signal consist of voiced part and unvoiced part. In conventional source-filter model, those two sources are considered to be independent. But in real situation it is difficult to segregate the source into voiced and unvoiced part. Actual source consist of mixture of two sources and the ratio varies according to the contents or intention of the speaker. In this paper we tried to segregate the components of voiced and unvoiced while considering source models. Source signals are modeled based on residual signal measured from inverse filtering. Two kinds of source models are assumed. Each model parameters are optimized to the original speech signal using genetic algorithm. The resulting parameters were compared in terms of the mel-cepstral distance to the original signal, spectrogram and spectral envelope from the synthesized signal.

Index Terms. Voice, source, model, synthesis, optimization

Full Paper (reprinted with permission from Firenze University Press)

Bibliographic reference.  Jo, Cheolwoo / Kim, Jaehee (2011): "Estimation of multiple source component using genetic algorithm", In MAVEBA-2011, 103-106.