Detection of Glottal Closure Instants From Speech Signals: A Quantitative Review

Abstract

The pseudo-periodicity of voiced speech can be exploited in several speech processing applications. This requires however that the precise locations of the Glottal Closure Instants (GCIs) are available. The focus of this paper is the evaluation of automatic methods for the detection of GCIs directly from the speech waveform. Five state-of-the-art GCI detection algorithms are compared using six different databases with contemporaneous electroglottographic recordings as ground truth, and containing many hours of speech by multiple speakers. The five techniques compared are the Hilbert Envelope-based detection (HE), the Zero Frequency Resonator-based method (ZFR), the Dynamic Programming Phase Slope Algorithm (DYPSA), the Speech Event Detection using the Residual Excitation And a Meanbased Signal (SEDREAMS) and the Yet Another GCI Algorithm (YAGA). The efficacy of these methods is first evaluated on clean speech, both in terms of reliabililty and accuracy. Their robustness to additive noise and to reverberation is also assessed. A further contribution of the paper is the evaluation of their performance on a concrete application of speech processing: the causal-anticausal decomposition of speech. It is shown that for clean speech, SEDREAMS and YAGA are the best performing techniques, both in terms of identification rate and accuracy. ZFR and SEDREAMS also show a superior robustness to additive noise and reverberation.

Improved automatic detection of creak

“…The first observation was the presence of secondary (or even tertiary) peaks proceeding the main excitation peak which corresponds to the glottal closure instant (GCI, Drugman et al, 2012c). This is illustrated in Figure 3, where strong residual peaks can be observed before the residual peak which corresponds to the GCI (as shown by the derivative of the EGG signal).…”

Tracheoesophageal speech: A dedicated objective acoustic assessment

“…In unvoiced speech, the excitation signal has a noisy nature and its kurtosis is relatively low (Drugman, 2014), while in voiced speech, it exhibits quasi-periodic discontinuities at the Glottal Closure Instants (GCIs, (Drugman et al, 2012b)) which is reflected by a greater sparsity and hence higher values of the kurtosis (Drugman, 2014). The kurtosis of the LP residual signal was used in (Falk et al, 2012) as an efficient way to characterize the vocal harshness in spastic dysarthric speech.…”

Speech polarity determination: A comparative evaluation

“…In [1], we proposed a method of Glottal Closure Instant (GCI) determination which relied on a mean-based signal. This latter signal had the property of oscillating at the local fundamental frequency and allowed good performance in terms of reliability (i.e leading to few misses or false alarms).…”

“…As in [1], the window length is recommended to be proportional to the mean period T 0,mean of the considered voice, so that y p 1 ,p 2 (t) is almost a sinusoid oscillating at the local fundamental frequency. For (p 1 , p 2 ) = (1, 1), the oscillating moment is the mean-based signal used in [1] for which the window length is 1.75 · T 0,mean .…”

“…For example, the three best approaches considered in [1] for the automatic detection of GCI locations, are dependent upon the speech polarity. An error on its determination results in a severe impact on the reliability and accuracy performance.…”