STUDY OF CORRELATION BETWEEN FACIAL ARTICULATOR MUSCLE MOTOR SIGNALS AND SPEECH SIGNALS IN VOCAL AND DIPHTHONG PHONEMES ARTICULATION

• Main Author: Rady Irawan, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50282
• Institution: Institut Teknologi Bandung

Information on muscle motor activity and speech has been used in various fields, such as technology, forensics, and health. In the health sector, muscle motor signals can also be used for various purposes, for example is to help quantify the progress of post-patient rehabilitation stroke in moving and speaking. The speech formation mechanism comes from a combination of motion in the speech system and muscle motor movements. The process of moving the speech system and motor muscle movement involves a lot of articulatory organs, one of which is the articulator organ around the oral cavity. This change in the shape of the articulator affects the resulting spoken sound, so that each phoneme has different characteristics of the articulation pattern. To determine the characteristics of the articulation pattern in the oral cavity on the speech formation of each phoneme, a correlation analysis can be used with the motoric muscle signals in the facial area measured using EMG. However, measuring the motor signals of the articulator muscles on the face has its own challenges to avoid the appearance of artifacts due to electrode motion. In addition, studies regarding the correlation of the two signals to determine the characteristics of the articulation pattern in the world are still very limited. So that to get the relationship between the two signals, the design of the instrument configuration for data collection, determining the measuring point for the placement of the electrodes, as well as the multifysis signal data retrieval protocol to minimize artifacts and simplify the data collection process is done first. Then perform signal processing for EMG and speech. In EMG signal processing, signal conditioning is first carried out using wavelet denoising 6 levels of with mother wavelet debauchies20 (20db) to remove background noise. The captured EMG signal features are envelope RMS, energy, and cross correlation. Meanwhile, in the speech signal processing, the feature short time energy (STE) and the third formant (F3) are taken. Furthermore, cross-correlation was performed on each pair envelopes of EMG signal to determine the tendency of the articulator shape in the speaking process. In addition, an analysis of the total contraction energy needed to move, hold and return the articulator position in the articulation processes of vowel and diphthong phonemes was carried out. Also, the change in the harmonic frequency (formant) of the spoken voice which is thought to have a correlation with motor signals in the process of forming the articulator is reviewed. From this research, the results show that the configuration of the acquisition instrument can be used to obtain muscle motor signals and speech sounds. In addition, the results show that the zygomaticus major and depressor anguli oris muscles are facial muscles that play a role in the articulation of vowels and diphthongs with the contribution of the muscles depressor anguli oris dominant. The second analysis of signal information using envelopes, cross-correlation, energy ratios and formant values can be used to analyze the character of muscle contraction patterns in the articulation process of vowel and diphthongs phonemes. The results also show that overall delay value of the cross-correlation results for the four pairs of electrodes has a relatively low value (0.024 ms and 0.136 ms for respondents 1 and 2). This indicates that the four points of the articulator muscle are moving relative to each other. This delay value indicates that each respondent has a tendency to have a unique contraction pattern in the articulator muscle and is influenced by various factors. In addition, the ratio of total lower articulator energy (zygomaticus major) for both respondents 0.374 and 0.034, which was higher than the upper articulator muscle (by 0.103 and 0.008 for respondents 1 and 2). This indicates that the lower articulator muscle plays a greater role in the articulation process of vocal phonemes and diphthongs phonemes. Vowel phonemes and diphthongs can be categorized based on the energy required in the process of forming, pronouncing and repositioning the articulator. Also, it was also found that there was an effect of changes in muscle contraction on the harmonic frequency (formant) of spoken sounds during the speaking process, especially during the diphthong phoneme transition. This research is a preliminary study, the findings and methods in this study can be used as a reference for developing the characterization of phoneme articulation based on the contraction behavior of the articulatory muscles.