Phoneme Synthesis
May 7th, 2009 by admin
The project “Inter-Voice Morphing using a neural network based dimensionality reduction technique” is created in collaboration with Tommaso Villanova. In this project we have attempted to set up the first experiments for a novel way to generate and morph voice sounds. In this project we use a special case of a feedforward neural network and train it using a large phoneme dataset. The idea is to encode the phoneme data in the neural network and extract its minimal amount of principal components. After training the neural network is used to construct a synthesizer for voice (phoneme) sounds.
For more information about the preliminary tests go here!