Ibon Oleagordia-Ruiz and Begonya García-Zapirain
Speech processing, Oesophageal voice, Discrete wavelet transform, Kalman filtering, Shimmer, HNR
This research work presents an oesophageal speech improvement algorithm using Wavelet and Kalman Filtering approach. In both techniques, it has been used different mother wavelet for wavelet approach and different measurement noises for Kalman filtering. People who have suffered from larynx cancer have enormously low intelligibility due to the surgery. A new algorithm has been developed to improve the speech quality. The algorithm consists of enhancing the Shimmer and HNR parameters using Discrete Wavelet Transform (DWT) and Kalman Filtering. By taking advantage of the Wavelet Transform's special time-frequency properties, a corrective algorithm in the form of a wave is applied for the signal intervals in which the shimmer measurement goes beyond normal levels. Therefore, an atomized control of the signal peaks is carried out, having an effect on the normalization of the shimmer. Regarding Kalman filtering, it is proved that the noise obtained from an oesophageal voice during periods of silence is the most suitable. The speech enhancement has been measured using Multidimensional Voice Program tool (MDVP)
Important Links:
Go Back
