Differences

This shows you the differences between two versions of the page.

Link to this comparison view

determination_and_selection_of_speaker_dependent_voice_features_for_vocal_tract_normalization [2014/03/15 12:49] (current)
bziolko created
Line 1: Line 1:
 +Mariusz Mąsior, Mariusz Ziółko (Poland)
 +
 +**Determination and selection of speaker dependent voice features for vocal tract normalization**
 +
 +Vocal Tract Normalization (VTN) is general concept concerning speech normalization to reduce the effects of differences between speakers vocal tracts. These techniques modify the speech signals in an attempt to reduce acoustic variability between speakers. VTN is widely used to improve the accuracy of automatic speech recognition.
 +
 +Typically, VTN concerns reduction of effects caused by different lengths of the human vocal tract. These effects differentiate the location of speech formant frequencies. Although this is a valuable parameter, the characteristic of speaker’s voice is determined by much more features (as speech rate, fundamental frequency, gender, prosody, emotions, etc.) which are usually not taken into consideration.
 +
 +This presentation introduces approach to speech analysis for identification of greater amount of speaker dependent voice features, which can help to expand voice normalization procedures. The adequate modelling of voice characteristics requires implementation of temporary voice features as well as long term parameters. Appropriate optimization and selection of introduced features leads to creation of consistent VTN algorithms’ parameterization.
  
Copyright © XXII PVC Organizing Committee 2013. All Rights Reserved.