Post-processing of Translated Speech by Pole Modification and Residual Enhancement to Improve Perceptual Quality

The perceptual quality of translated speech depends on the quantity of speech data used for training. The translation speech quality is poor when the system is trained with less data. The quality improves by gradually adding more speech data for training. This work demonstrates the significance of post-processing of translated speech by signal processing for improving perceptual quality. Initially, the target speech original residual is used to replace the translated speech residual. It is then replaced using the weighted residual obtained by speech enhancement. The pole modification of translated speech is also done. Finally, both weighted residual and pole modifications are combined. All the experiments show improvement in perceptual quality.