IntroductionThe goals of this study are 1- to describe precisely the muscle coordination during French bilabial stop consonant production, in both time and amplitude, using surface electromyography and 2- to find relevant and reliable descriptors of speech articulation effort. Material and methodsAn experiment in two parts was conducted. A first methodological part aimed at targeting multiple orofacial muscles, at validating the correct position of EMG electrodes on the speaker’s face, and at ensuring that the recorded EMG signals were not significantly contaminated by the activity of adjacent muscles. A second part aimed at characterizing the time activation patterns of these orofacial muscles during the production of labial stop consonants, at examining how these patterns were influenced by the following vowel and by an increasing level of articulation effort, and at finding muscle activation descriptors that reproducibly correlate with kinematic parameters such as the degree of interlip compression or articulatory velocities.To that goal, four adult speakers without speech disorders were recorded in laboratory conditions. In the first part of the experiment, they were asked to produce 6 non-speech orofacial movements (lip spreading; lip compression; lowering of respectively the jaw and the lower lip, inDLIendently; lip protrusion and raising of the lower lip), with 20 repetitions of each movement. In the second part of the experiment, they produced 6 speech logatoms /lepa/, /leba/, /lepi/, /lebi/, /lepu/ and /lebu/ by series of 5 items, with an increasing level of articulatory effort (self-evaluated, from 1 to 5). Ten repetitions of these series were recorded for each logatom.Two types of signals were simultaneously recorded: high-speed images of the lips (at 100Hz) and the EMG signal of 7 different orofacial muscles recorded at 20 kHz with bipolar surface electrodes (at 200Hz, with a Biopac system). The seven pairs of surface electrodes were placed on the superior orbicularis oris (OOS), the inferior orbicularis oris (OOI), the zygomatic (ZYG), the buccinator (BUC), the depressor labii inferioris (DLI), the mentalis (MNT) and the anterior belly of the digastric (DIG). The internal and external lip contours were extracted from the high-speed images. For the non-speech gestures of the first experimental part, two time intervals were segmented from the kinematic signals, corresponding to P1- the lip displacement from the rest position to their maximum displacement and P2- the back displacement to the rest position. For the speech gestures of the second experimental part, four time intervals were segmented from the kinematic signals, corresponding to P1- lip closing, P2- lip compression increase, P3- lip compression decrease and P4- lip re-opening. Three kinematic parameters were extracted from the external lip contour: the degree of interlip compression, the articulatory velocities of lip closing and re-opening.For both experimental parts, muscle activity was analyzed considering the integral of the full-wave rectified EMG signal over each time interval of the gestures. Two different methods were used to investigate the degree of co-activation of the different muscles in each phase of the movements (correlation between muscle activity levels) and the degree of similarity of the recorded EMG signals in each phase of the movements (raw inter-correlation between full-wave rectified EMG signals envelopes).ResultsPreliminary results on one speaker suggest that a correct position was found for all pairs of EMG electrodes but one (targeting the OOS muscle), ensuring that these electrodes recorded the activity of the targeted muscles. High levels of co-activation were observed between some muscles in several phases of non-speech gestures (R2 > 0.65). However, for most of the electrodes, these high levels of co-activations were not accompanied by a significant degree of similarity between the full-wave rectified EMG signals envelopes (R2 < 0.40), ruling out any problem of diaphony between these adjacent electrodes. On the contrary, when the DLI muscle was activated, the EMG signals from the MNT and the DLI electrodes always showed a high level of co-activation and a high degree of waveform similarity, which supported the existence of a diaphony phenomenon from the DLI electrode to the MNT one. As a result, the signal from the OOS and MNT electrodes was no longer considered to study the speech production gestures.Furthermore, results obtained from the second experimental part showed that the production of bilabial stop consonants could be characterized by a reproducible time activation pattern of orofacial muscles: the first phase of the gesture (P1-lip closing) was characterized by a maximal activity of the BUC muscle, while the third phase (P3-lip compression decrease) was underlined by a significant and maximal activity of the DLI, DIG and OOI muscles. This time activation pattern was not significantly influenced by the following vowel or by the level of articulation effort. On the other hand, the amplitude of these muscle activities was overall significantly greater in /a/ and /i/ contexts, compared to a /u/ context (except for the OOI muscle that showed, on the contrary, a greater activity in the /u/ context). These muscle activities were also highly affected by the level of articulation effort. In particular, a significant correlation was observed between the level of articulation effort and the activity of the BUC muscle in P1 (R2 = 0.72), and of the DLI and DIG muscles in P3 (R2 = 0.70 and R2 = 0.71). Finally, significant correlations were also found between these muscle activities and kinematic descriptors of the lip gesture. In particular, the activity of the DIG muscle in P3 highly correlated with the degree of interlip compression (R2 = 0.71) and with the lip opening velocity (R2 = 0.75), while the activity of the BUC muscle in P1 correlated significantly with the lip closing velocity (R2 = 0.72).ConclusionThis study brings a methodological contribution to the investigation of orofacial muscle activities in speech production, using surface electromyography, by exploring in detail the existence of signal crosstalk between adjacent pairs of electrodes. The observations provided here support the interest of decomposing the global speech gesture into phases in order to characterize more precisely the sequencing of the activations and the co-activations across muscles. Finally, reproducible descriptors of speech articulation effort were found for a first speaker in the activity of 2 muscles (BUC and DIG), which correlated significantly with both the sensation of effort and with kinematic descriptors. Results from the 4 speakers will be presented and discussed during the conference.