PROJECT TITLE :
Instantaneous model adaptation method for reverberant speech recognition
An acoustic model adaptation algorithm is proposed for reverberant speech recognition. Inspired by the eigenvoice adaptation framework, multiple acoustic models reflecting varied reverberant environments are combined for instantaneous adaptation. Using artificially generated reverberant speech, multiple acoustic models are trained consistent with multiple reverberation times. The mean vectors of the optimal acoustic model are obtained as a weighted sum of these of multiple acoustic models by employing a most-chance criterion. For effective model combination, reverberant speech is preprocessed. Experiments on English continuous speech recognition tasks during a simulated reverberant surroundings show that the proposed method performs higher than the conventional adaptation techniques.
Did you like this research project?
To get this research project Guidelines, Training and Code... Click Here