Speech Prosody 2010

Chicago, IL, USA
May 10-14, 2010

A New Bidirectional Neural Network Model for the Acoustic- Articulatory Inversion Mapping For Speech Recognition

Hossein Behbood, Seyyed Ali Seyyedsalehi, Hamid Reza Tohidypour

Amirkabir University of Technology, Dep. of Biomedical Engineering, Tehran, Iran

In this paper, a new bidirectional neural network for better acoustic-articulatory inversion mapping is proposed. The model is motivated by the parallel structure of human brain, processing information by having forward-inverse connections. In other words, there would be a feedback from articulatory system to the acoustic signals emitted from that organ. Inspired by this mechanism, a new bidirectional model is developed to map speech representations to the articulatory features. In comparison with a standard model, the output of bidirectional model as auxiliary data in phone recognition process, increases the accuracy up to approximately 3%.

Index Terms: Bidirectional Neural Networks (BNNs), Feed-Forward Networks (FFNs), Time Delay Neural Networks (TDNNs), MOCHA-TIMIT database, Acoustic-articulatory inversion mapping

Full Paper

Bibliographic reference.  Behbood, Hossein / Seyyedsalehi, Seyyed Ali / Tohidypour, Hamid Reza (2010): "A new bidirectional neural network model for the acoustic- articulatory inversion mapping for speech recognition", In SP-2010, paper 580.