Accurate estimation of glottal closure instants (GCIs) in voicedspeech is important for speech analysis applications which benefitfrom glottal-synchronous processing. Electroglottograph (EGG)recordings give a measure of the electrical conductance of the glottis,providing a signal which is proportional to its contact area. EGGsignals contain little noise or distortion, providing a good referencefrom which GCIs can be extracted to evaluate GCI estimationfrom speech recordings. Many approaches impose a threshold onthe differentiated EGG signal which provide accurate results duringvoiced speech but are prone to errors at the onset and end ofvoicing; modern algorithms use a similar approach across multipledyadic scales using the stationary wavelet transform. This paperdescribes a new method for EGG-based GCI estimation namedSIGMA, which is based upon the stationary wavelet transform, peakdetection with a group delay function and Gaussian Mixture Modellingfor discrimination between true and false GCI candidates.In most real-world environments, it is necessary to estimateGCIs from a speech signal recorded with a microphone placedat some distance from the talker. The presence of reverberation,noise and filtering by the vocal tract render GCI detection fromreal speech signals relatively difficult to achieve compared with theEGG, so EGG-based references have often been used to evaluateGCI detection from speech signals. Evaluation against 500 handlabelledsentences has shown an accuracy of 99.35%, a 4.7% improvementover a popular existing method.
展开▼
