New approach for oesophageal speech enhancement

摘要

This paper presents an oesophageal speech enhancement algorithm. Such an exceptionally special type of voice is due to the laryngectomy undergone by those persons with larynx cancer. An oesophageal voice has extremely low intelligibility. This work proposes a method to improve its quality, which consists of stabilizing the transfer function poles of the vocal tract model so as to improve a signal's formants. With this aim, the joint use of three techniques has been applied: firstly, Digital Wavelet Transform technique, secondly, Kaiman filtering technique and thirdly, an algorithm that transforms the modulus and phase of vocal tract's poles. When using this efficient speech enhancement Kaiman Filtering procedure, the speech signal is usually modelled as an autoregressive (AR) process and represented in the state-space domain. The final speech improvement has been measured with the help of Multi-Dimensional Voice Program (MDVP) tools.