Towards Evaluating Pitch-Related Phonation Function in Speech Communication Using High-Density Surface Electromyography

Abstract

Pitch, as a sensation of the sound frequency, is a crucial attribute toward constructing a natural voice for communication. Producing intelligible sounds with normal pitches depend on substantive interdependencies among facial and neck muscles. Clarifying the interrelations between the pitches and the corresponding muscular activities would be helpful for evaluating the pitch-related phonating functions, which would play a significant role both in training pronunciation and in assessing dysphonia. In this study, the speech signals and the high-density surface electromyography (HD sEMG) signals were synchronously acquired when phonating [a:], [i:], and [ә:] vowels with increasing pitches, respectively. The HD sEMG energy maps were constructed based on the root mean square values to visualize spatiotemporal characteristics of facial and neck muscle activities. Normalized median frequency (nMF) and root-mean square (nRMS) were correspondingly extracted from the speech and sEMG recordings to quantitatively investigate the correlations between sound frequencies and myoelectric characteristics. The results showed that the frame-wise energy maps built from sEMG recordings presented that the muscle contraction strength increased monotonously across pitch-rising, with left-right symmetrical distribution for the face/neck. Furthermore, the nRMS increased at a similar rate to the nMF when there were rising pitches, and the two parameters had a significant correlation across different vowel tasks [(a:) (0.88 ± 0.04), (i:) (0.89 ± 0.04), and (ә:) (0.87 ± 0.05)]. These findings suggested the possibility of utilizing muscle contraction patterns as a reference for evaluating pitch-related phonation functions. The proposed method could open a new window for developing a clinical approach for assessing the muscular functions of dysphonia.

Keywords: high-density; phonation function; pitches; speech communication; surface electromyogram.

Figure 1

(A) Experimental setup to record the speech and HD sEMG signals simultaneously; (B) Placement of the high-density surface electrodes on the facial and neck muscles.

Figure 2

The amplitude and spectrogram of the speech signals when phonating the vowel [a:] with continuously increasing pitch. (A) The amplitude of the speech signal as a function of time; (B) The time-frequency spectrogram of the speech signal.

Figure 3

Amplitude spectra of the speech signal with an increasing pitch from a series of analysis windows (F1–F40).

Figure 4

The temporal waveforms of the HD sEMG signals recorded from all the 120 surface electrodes on the facial and neck muscles during the phonation with rising pitch.

Figure 5

The HD sEMG energy maps of the facial and neck muscles by using 40 analysis windows with a length of 250 ms during phonating vowel [a:] with pitch increase: (A) A typical frame of HD sEMG energy map visualizing the facial and neck muscle activities; (B) HD energy maps of the facial muscles; (C) HD energy maps of the neck muscles.

Figure 6

(A) Comparison of the nMF of the speech signals and the nRMS of the sEMG recordings from one channel when phonating vowel [a:] with increasing pitch; (B) Correlation coefficients between the nRMS values of 120 channels of sEMG recordings and the nMF of the speech signal from three repeated tests.

Figure 7

The averaged HD sEMG energy maps of the facial and neck muscles during the phonating vowel [i:] with a pitch increase.

Figure 8

The averaged HD sEMG energy maps of the facial and neck muscles during the phonating vowel [ә:] on a rising pitch scale.

Figure 9

(A) Comparison of the nMF of the speech signals and the nRMS of the sEMG recordings from channel E8 when phonating three different vowels [(a:), (i:), and (ә:)] with increasing pitch; (B) Correlation coefficients between the nRMS values of 120 channels of sEMG recordings and the nMF of the speech signal across three different vowels [(a:), (i:), and (ә:)].