Publication - Reducing the cocktail party effect prior to Automatic Speech recognition
PROFILE

Reducing the cocktail party effect prior to Automatic Speech recognition

Research Area:  
    
Type:  
In Proceedings

 

Year: 2009
Authors: Theologos Athanaselis; Stylianos Bakamidis; Ioannis Dologlou
Book title: Proceedings of the 13-th International Conference Speech and Computer (SPECOM)
Pages: 360-363
Abstract:
This work deals with signals impaired by white noise, coloured noise and cocktail party effect with varying Signal-to-Noise Ratio. A non-linear spectral subtraction method (NSS) will be used in conjunction with the Large Vocabulary Continuous Speech Recognition system developed for the HEARCOM project (004171) to quantify the impact of speech enhancement on speech recognition performance. It is shown that the impact of NSS depends on the type of noise.
[Bibtex]