Abstract
Vibrato enhances the expressiveness of singing, but mastering and controlling it requires advanced vocal skills. If vibrato is controlled through engineering methods, even beginners can achieve high expressiveness. This study develops a system that controls vibrato by applying electrical stimulation to the larynx. This approach enables motor-level control of vocal modulation, linking users’ physical movements to interactions with virtual reality avatars in virtual environments. Experiments were performed to investigate several aspects: whether the extent of vibrato changes depending on stimulation position and it can be controlled by current intensity, whether vibrato rate can be controlled by stimulus frequency, and whether variations in fundamental frequency affect vibrato extent. The results indicate that stimulation position influences vibrato extent. Additionally, vibrato extent significantly differs based on current intensity. The median vibrato rate aligned with the stimulus frequency for eight out of eleven participants, indicating a clear observation of vibrato. A significant correlation was observed between the fundamental frequency of vocalization and vibrato extent for three participants, suggesting that pitch affects vibrato parameters in some cases. These results indicate the feasibility of controlling vibrato rate and extent through adjustment of the stimulation parameters. This system shows the potential for flexible modulation of vibrato extent and rate according to the song, thereby augmenting human singing ability.
1 Introduction
Voice is one of the most familiar sounds and vibrations to humans. In addition, voice is a fundamental modality through which users interact with their environment. Its modulation has been discussed in the context of virtual reality as a component of embodied interaction, particularly in relation to avatar ownership. In virtual environments, users’ voices are often perceived as part of the virtual self through avatars, making vocal control a potentially important yet underexplored modality of embodiment. For example, synchrony between audiovisual stimuli and vibrations applied to the throat has been shown to affect the sense of embodiment (; ). Prior studies suggest that constructing an interaction loop that intervenes in the human motor loop can influence embodiment and immersion in virtual reality avatars.
Singing, a form of vocalization, is a technique for self-expression and communication. The demand for singing is increasing in education () and medical contexts (). Vibrato, a technique used to enhance singing expression, plays a key role in this context. According to , “the vibrato physically corresponds to a periodic, rather sinusoidal, modulation of the phonation frequency.” Vibrato adds “richness” to timbre (), enhances voice segregation (), and positively influences singing evaluations (). claimed that vibrato introduces pleasing flexibility, mellowness, and tonal richness. reported that voices with vibrato were more segregated than those without it. Previous studies have also shown that appropriate vibrato influences singing evaluations ().
Vibrato is mainly characterized by two features: vibrato extent (amplitude of vibrato) and vibrato rate (modulation frequency). The vibrato extent typically ranges from 20 to 150 cents (; ; ), while the vibrato rate generally falls between 4–8 Hz in professional singers (; ). In particular, a vibrato rate of 5–6 Hz tends to receive high evaluations (; ). Although humans struggle to produce vibrato at 8 Hz or higher ; ; , such rates can expand the expressive range and can be accepted as a unique singing style that exceeds typical physiological limits.
However, mastering vibrato is generally challenging. revealed that after 9 h of singing instruction, three amateur male singers showed increased vibrato extent comparable to that of professional singers, along with reduced decreased. However, the vibrato rate (modulation frequency) did not approach professional levels, and its variance also did not change. Receiving singing lessons from a professional for two to 3 years improved the vibrato rate and reduced its variance (; ). Significant time is required to learn to use vibrato appropriately, and controlling the vibrato rate remains difficult. Additionally, independently applying vibrato to match target characteristics remains challenging (; ).
Therefore, this study aimed to develop a system that can generate vibrato regardless of the user’s skill level. In this study, we focused on controlling vibrato through external intervention using electrical stimulation of the larynx. By controlling the vibrato extent and rate, the proposed system enables a novel form of vocal expression that is difficult to produce without assistance.
Moreover, vibrato modulation can influence sensorimotor interactions between physical movement and the environment. This approach is closely related to movement modulation techniques employed in virtual reality, such as transparent-body perception () and perceptual modulation through electrical muscle stimulation (). While such approaches are well established for body movement through audiovisual or haptic technologies, vocal interaction in virtual reality remains limited to passive audio capture, followed by post hoc signal processing or vibration feedback. Consequently, vocal motor control is typically excluded from the embodied interaction loop. Beyond its role in sensory feedback, vibrato modulation may also allow direct control of vocal motor output. This perspective positions vocal motor control as a modality that can potentially function beyond passive signal processing in virtual environments. This capability could potentially be applied to virtual reality systems by incorporating vocal production into avatar interaction loops, although such applications remain to be validated in future work.
Several studies have shown that electrical stimulation of the larynx alters the fundamental frequency (F0) of speech. This effect has been clearly demonstrated by invasive research. For example, stimulated the cricothyroid and thyroarytenoid muscles in five participants using needle electrodes and found that both caused an increase in F0. In non-invasive studies, and applied 1 hour of surface electrical stimulation of the larynx in 30 and 10 participants, respectively, and examined changes in F0 before and after stimulation. Although no significant group-level differences were found, the direction and magnitude of the F0 changes differed significantly across participants. monitored one participant’s voice every 3 minutes during a 12-min surface electrical stimulation session and observed a significant increase in F0 during and after stimulation.
Studies have also investigated the application of vibrato through electrical stimulation to the laryngeal region. As an invasive method, Titze et al. applied stimulation by inserting needle electrodes into the cricothyroid muscle. As a result, the vibrato modulation extent increased, and particularly in trained singers, a clear peak in modulation extent was observed around 5 Hz (). In contrast, as a non-invasive approach, Inbar et al. conducted experiments using surface electrodes arranged to primarily stimulate the suprahyoid muscle group. Their results confirmed that the first formant (the lowest resonance frequency of the vocal tract) was modulated in synchronization with the stimulation frequency of 5 Hz ().
To apply an ideal vibrato, we must verify whether it can be controlled by electrical stimulation of the larynx. Previous research has suggested that electrical stimulation of the larynx can induce perceptual changes in vibrato extent and rate, as demonstrated in user studies (). However, these studies are not sufficient to evaluate the relationship between stimulation parameters, such as stimulation position, and the resulting sound waveform. The larynx contains numerous muscles, and muscle activation varies significantly depending on the stimulation site. In addition, certain muscle groups may cause large fluctuations in the fundamental frequency (F0), which is essential for effective vibrato. Therefore, it is crucial to evaluate the effects of selective electrical stimulation.
Accordingly, this study aimed to develop a selective electrical stimulation for the laryngeal muscles to control vibrato. Four experiments were conducted. First, Experiment 1 focused on the electrode position. Vibrato can be applied more efficiently by identifying a position producing a high vibrato extent with low current intensity. Next, Experiment 2 examined the effect of current intensity. Third, Experiment 3 investigated whether stimulus frequency affects vibrato characteristics. Fourth, Experiment 4 explored whether changes in F0, which involve vertical laryngeal movement, influence vibrato characteristics. By clarifying the relationships between vibrato characteristics, stimulation parameters, and F0, the developed system can achieve ideal vibrato control.
2 Materials and methods
2.1 Experimental setup
We developed a selective electrical stimulation system to determine the optimal electrode position and control the vibrato extent and rate. The system consists of a microphone, a module for measuring and analyzing singing sounds, and a selective electrical stimulator that receives commands from a PC and delivers stimulation to participants (Figure 1).
FIGURE 1
The selective electrical stimulator consists of a microcontroller, stimulator circuit, electrode selector circuit, and an electrode array (Figures 1A,B) (). The stimulator operates at a constant current. A biphasic square waveform is applied (Figure 1C), and the current intensity, frequency, and electrode position are adjusted during the experiments. For safety, the current intensity is limited to 10 mA using current-limiting diodes. The pulse width of the stimulation is 0.5 m. The electrode selector uses an analog switch. The electrodes are fabricated on a flexible substrate with 64 channels ( channels used as anodes and cathodes). Each electrode measures mm, with 2 mm spacing between electrodes. One electrode is used as the anode and another as the cathode, positioned with a 10 mm gap at the center. The anode and cathode are arranged symmetrically with three electrode spacing configurations. Electrodes are arranged in a grid, forming 21 electrode positioning patterns (Figure 1D). A mm gel pad is attached to each electrode, leaving a gap between them to prevent overlapping, and then applied to the skin. The charge density is up to 5 . The actual current flow path depends on the skin’s resistance distribution and the electrical properties of the underlying volume.
Sound was measured using a microphone (ECM8000, Behringer, sampling rate: 48 kHz). The sound data were analyzed using a computer. Instructions were provided to participants via a display and speakers, and they were instructed to proceed with the experiment using a mouse. Participants could also stop the electrical stimulation at any time by clicking a designated area on the screen.
2.2 Participants
Fourteen healthy adults (nine males and five females) participated in this experiment. Data from 11 participants (seven males and four females) were included in the analysis, as three participants (two males and one female) were excluded due to insufficient current intensity during the protocol described below. All experiments were approved by the Life Science Research Ethics and Safety Office of the University of Tokyo, Japan (approval number: 23–350). All participants signed an informed consent form after being provided with an overview of the experiment and instructions. The study protocol was conducted in accordance with the ethical standards of the Declaration of Helsinki. No adverse events were observed throughout the experiment.
2.3 Preparation
Figure 2A shows the experimental setup. First, participants were asked to sit on a chair with no backrest and instructed to lean their backs against the wall. A headrest was attached to the wall to support the participant’s head. They were then asked to relax, speak at a comfortable volume, and maintain a consistent volume throughout the trials. Participants were instructed to vocalize the vowel “/a/” during the experiment. Next, they were instructed to attach the electrodes as shown in Figure 2B. Before attaching the electrodes, participants wiped the area around the larynx and positioned the electrodes at the center of the neck. Using a mirror, they ensured that the electrodes did not shift even when they lowered their chin. The upper and lower parts of the electrodes were secured with surgical tape. A case containing the electrode selector circuit was attached to their clothing to prevent tension from accumulating in the cords.
FIGURE 2
After setting up the device, participants proceeded with the experiment by looking at the display and operating the mouse. First, they selected a pitch for the experiment, corresponding to a single note from the A, B, C, D, E, F, and G scales, and participants could choose any octave they preferred to maintain a consistent F0 throughout the experiments, except in Experiment 4. In the following experiments, participants listened to the selected pitch for one second before vocalizing. After selecting the pitch, they recorded 3 s vocalizations without any stimulation. Through this procedure, participants practiced the sequence of actions in which they pressed a button and then produced vocalization. These data were also analyzed in the same manner as the experimental data, and the vibrato extent was calculated. Next, the participants became accustomed to the stimulation by gradually increasing its intensity. The current was increased in 17 steps from 2 to 10 mA in 0.5 mA increments.
Stimulation was applied for 1 s at each current intensity across 21 different electrode positions. The electrode positions were stimulated in a randomized order using 21 combinations of left- and right-symmetrical electrode sets (Figure 1D). Before each increase in intensity, the participants took a 10 s break. Positions at which participants reported pain were excluded from subsequent experiments. In addition, if participants reported pain at more than eight positions, the current intensity was not increased further. At this stage, three participants (two males and one female) reported pain. For these participants, the experiment was conducted at the highest pain-free current intensity; however, their data were excluded from the analysis. The maximum current intensities for the three participants were 2.5 mA, 3 mA, and 6.5 mA, respectively.
2.4 Methods
Figure 3 shows the timeline of the experiments. In Experiment 1, the effect of changes in electrode position on the vibrato extent was investigated. Participants listened to the tone of the scale selected during the preparation phase for 1 s. After the button press, stimulations were applied up to five electrode positions in sequence for 1 s each, beginning at the onset of vocalization (Figure 3A). Because participants could self-pace the timing of vocalization onset, they had sufficient time to prepare before stimulation began. These trials were repeated until 1 s of data was obtained for each electrode position, excluding those that had caused pain during preparation. The electrode positions are presented randomly. The current intensity was set at 10 mA, and the frequency was 6 Hz. The vibrato extent was calculated for each electrode position. The position with the maximum vibrato extent was used in subsequent experiments.
FIGURE 3
Experiment 2 focused on the relationship between current intensity and the vibrato extent. As in Experiment 1, participants listened to the scale tone selected during the preparation phase for 1 s. Next, they received stimulation at 6 Hz with current intensities of 4, 6, 8, and 10 mA, starting 1 s after the beginning of vocalization. Each participant completed eight trials per session: two trials for each current intensity, presented in a random order. The experiment consisted of five sessions, totaling 40 trials. Voice was recorded for 3 s (Figure 3B).
In Experiment 3, we focused on the vibrato rate and how it changed with stimulus frequency. Participants received stimulation at 10 mA and frequencies of 4, 6, 8, and 10 Hz, starting 1 s after the beginning of the vocalization. As in Experiment 2, 10 trials were conducted for each frequency (40 trials in total) across five sessions (Figure 3B).
In Experiment 4, we focused on the vibrato rate when F0 was changed. We presented the tones of (A, B, C, D) and (E, F, G, H) for increasing pitch, and (H, G, F, E) and (D, C, B, A) for decreasing pitch. Participants were instructed to produce the same pitch as presented. Stimuli were then applied during vocalization. The current intensity was set at 10 mA, and the frequency was set at 6 Hz. A total of 16 datasets were obtained. Each pitch was recorded for 1.5 s (Figure 3C).
2.5 Data analysis
Figure 4 presents an overview of data analysis. First, the data were sampled at 48 kHz, and audio was obtained in more than 80% of the sections. The raw data were converted to F0 data using Harvest () with a 1 m frame shift. The data were then trimmed to 1.0 s in Experiment 1 for each electrode position and 3.0 s in Experiments 2 and 3 for each trial. In experiment 4, the data were then trimmed to a 1.0 s segment centered within a period of continuous data lasting more than 1 s. Data without continuous sections of at least 1 s were excluded (one trial).
FIGURE 4
The vibrato rate was calculated using the F0 spectrum. The F0 spectrum was computed by applying a fast Fourier transform (FFT) to the trimmed data. The frequency resolution of the FFT is determined by the recording duration (in seconds), and is given by Equations 1, 2:where is the sampling frequency and is the number of samples. Intuitively, a longer recording duration yields finer frequency resolution: = 3.0 s gives = 0.33 Hz, while = 1.0 s gives = 1.00 Hz. The vibrato rate was then calculated from the frequency with the maximum amplitude in the 3–30 Hz range of the F0 spectrum, as human innate fluctuations occur primarily below 3 Hz.
The vibrato extent was calculated from the low-pass filtered (LPF) data. The cutoff frequency of the LPF was set to 2.5 the stimulus frequency (that is, 15 Hz in Experiments 1, 2, and 4; in Experiment 3, it was based on the stimulus frequency). We then calculated the vibrato extent from the LPF data using the method proposed by . First, we converted the filtered F0 data into F0 cent data. The vibrato extent was derived from the median of the differences between the linearly interpolated frequencies at troughs and peaks (, Figure 5A). The method used to derive the vibrato extent was as follows, as shown in Equations 3, 4:
FIGURE 5
To reduce the effect of outliers, we used the median instead of the mean, in contrast to the method used by . In addition, if the next stimulus arrives before the contraction of the muscle owing to electrical stimulation is completed, the vibrato extent may decrease (Figure 5C) (). To consider this effect, we simulated the vibrato extent at 6, 8, and 10 Hz by extrapolating from 4 Hz data, using recordings from participants whose stimulus and vibrato frequencies matched those in Experiment 3. First, we converted F0 to force as follows, referring to , as shown in Equation 5.:
The values of the various constant parameters were set to ρ = 1,140 kg/m3, S = 40.9 mm2, and L0 = 18.3, based on the biomechanical constants of the thyroarytenoid muscle reported in . However, this did not affect the F0 simulation because the values were returned to the F0 trajectory. Next, we extracted three parameters from the data, following , illustrated in Figure 5B: the maximal force caused by one stimulus , the time from contraction onset to peak force , and the time from contraction onset to when the force declined to half of . In this simulation, we assumed that F0 shifted downward owing to stimulation. All parameter values were based on median data. Trials in which the amplitude failed to reach half of after peaking were excluded because could not be determined. Subsequently, we then derived the waveform as follows (), as shown in Equations 6–9:
The simulated waveform was then generated by summing the individual waveforms corresponding to each stimulation pulse at the given stimulus frequency , as follows (Figure 5C), as shown in Equations 10, 11:
Finally, the simulated force was converted back to F0, as described in Equation 12, and the vibrato extent was calculated using the same method as that in Figure 5A.
All data were analyzed using the “rstatix” package in R (Version 4.4.3). In Experiments 2 and 3, the Wilcoxon signed-rank test, a nonparametric test, was used to assess statistical significance. A correlation analysis was conducted in Experiment 4. The statistical significance was defined as a two-sided p-value 0.05.
3 Results
As mentioned above, the analysis was conducted on 11 participants (seven males and four females) who were able to increase the current intensity to 10 mA. Participants p1-p6 and p11 were males, and p7-p10 were females.
First, we analyzed voice data without electrical stimulation. In the frequency range above 3 Hz targeted in this study, an average fluctuation of 5.95 cents was observed, and no participants exhibited vibrato frequencies exceeding 10 cents. Six participants showed fluctuations exceeding 5 cents: p1, p3, p5, p7, and p8 exhibited fluctuations in the 3–4.67 Hz range, while p11 showed fluctuations in the 4–8.3 Hz range.
3.1 Experiment 1: electrode positions
Figure 6A presents a heat map of the vibrato extent at each stimulation position. The positions where the vibrato extent was highest are squared in yellow. Positions that caused pain and were therefore excluded from measurements are marked as 0. Figure 6C shows the correspondence between these results and actual electrode locations. In the graph, a longer distance between the anode and cathode is represented on the left side. The heat map shows that the vibrato extent varied depending on the stimulation position. In particular, the vibrato extent exceeded 20 cents at certain positions for p1-p7, p9, and p11. For p1, the highest vibrato extent was observed at the upper-left position, bottom left for p2 and p6, and the second or third row from the bottom for the other participants.
FIGURE 6
Figure 6B shows a heat map depicting the vibrato rate for each stimulation position. Similar to Figure 6A, the positions with the highest vibrato extent are marked with yellow squares, and positions that caused pain and were not measured are indicated as 0. For p1–p9, the positions at which the vibrato rate aligned with the stimulus frequency (6 Hz) tended to cluster together. In p5 and p7, the positions where pain occurred were next to those where the vibrato extent peaked. In addition, for p2-p7 and p9, the stimulation positions associated with the highest vibrato extent also produced vibrato rates near the stimulus frequency of 6 Hz. However, this pattern was not observed for p1, p8, p10, or p11.
3.2 Experiment 2: current intensity
Figure 7A shows a box plot of the vibrato extent at various current intensities. The median vibrato extent was 5.74, 7.28, 18.21, and 23.55 cents for stimulation currents of 4, 6, 8, and 10 mA, respectively. Significant differences were found in multiple comparisons using the Wilcoxon signed-rank test with Holm correction. These differences include: 4–8 mA , 4–10 mA , 6–8 mA , 6–10 mA , and 8–10 mA . Asterisks in the figure indicate the significant differences between the current intensities. For some participants, the median vibrato rate matched a stimulus frequency of 6 Hz as the current intensity increased. Specifically, the vibrato rates for p4, p6, and p9 matched the stimulus frequency at 4 mA, while p2, p3, and p5 matched it at 6 mA, and p1 and p7 matched it at 10 mA.
FIGURE 7
3.3 Experiment 3: stimuli frequency
Figure 8B shows the median vibrato rate as a function of stimulus frequency. The vibrato rates for p1-p7, and p9 matched the stimulus frequencies, whereas p8, p10, and p11 did not match. Figure 8A shows a box plot of the vibrato extent for different stimulus frequencies. Vibrato extent decreased with increasing stimulus frequency, with significant differences identified by Wilcoxon signed-rank test with Holm correction between 6 and 10 mA and between 8 and 10 mA . Asterisks indicate significant differences between stimulus frequencies. Figure 8C shows a boxplot of the vibrato extent for each participant, alongside the simulated vibrato extent for 6, 8, and 10 Hz data derived from 4 Hz based on participants whose stimulus frequency and vibrato rate matched (i.e., p1–7 and p9).
FIGURE 8
3.4 Experiment 4: F0 pitch
Figure 9A shows the vibrato extent relative to F0 for each participant. One data point was excluded owing to the inability to calculate the vibrato extent because no two or more maximum points were identified in the filtered F0 waveform. The median F0 values from Experiment 3 are indicated by the black lines on the graph. The results of the correlation test revealed significant correlations for p3 , p6 , and p9 . Figure 9B shows the vibrato rate relative to F0. This figure indicates that the vibrato rates for p2, p6, and p9 matched the stimulus frequency (6 Hz) for all data. For p3 and p5, all data except for one matched the stimulus frequency (6 Hz).
FIGURE 9
4 Discussion
4.1 Discussion
4.1.1 Experiment 1: electrode positions
Figure 6A shows that the vibrato extent varied depending on the position of the stimuli. This variation highlights the need to adjust the electrode position according to the stimulation position. This is likely because the muscle groups are stimulated, and their activity levels differ depending on the position; therefore, the degree of elongation of the vocal cords also changes. In particular, the vibrato extent exceeded 20 cents for p1-p7, p9, and p11. The vibrato extent typically used is approximately 20–150 cents (; ; ), suggesting it to be a cognitively natural range for these participants.
Note that the observed vibrato cannot be solely attributed to electrical stimulation. As shown in Figure 6B, there were multiple positions where the stimulus frequency (6 Hz) and vibrato rate matched for p1-p9. In contrast, p10 and p11 showed only one or fewer matching positions. If the vibrato is solely caused by electrical stimulation, the vibrato rate should match the stimulus frequency. Therefore, the natural vocal fluctuation contributed more strongly at these positions (p10 and p11) than the electrical stimulation. In addition, different muscles were stimulated in each participant. The vibrato extent was greatest at the uppermost left for p1, bottom left for p2, and second or third rows from the bottom for the others.
Figure 6B shows the positions where the stimulus frequency (6 Hz) and vibrato rate tended to cluster together, suggesting that the electrode size can be extended to areas where the stimulus frequency and vibrato rate match. However, for p5 and p7, pain occurred next to the positions with the maximum vibrato extent. Although the vibrato rate can be controlled using a larger electrode, this could inadvertently include positions that cause pain. Additionally, different positions could stimulate different muscle groups. Specifically, at p1 and p8, the stimulus and vibrato frequencies matched. When a larger area is stimulated, muscle groups that control F0 in both upward and downward directions may be activated simultaneously, potentially decreasing the vibrato extent. However, selectively stimulating muscle groups that control F0 in one direction could enhance the vibrato extent. Therefore, selecting an appropriate electrode position and size is crucial, considering individual muscle structure and the potential for causing discomfort in each participant.
In p1 and p8, positions where the 6 Hz vibrato was observed were concentrated in the lower and upper regions, respectively. However, the maximum vibrato extent did not occur at this position. Therefore, positions where the vibrato rate did not match the stimulus in Experiments 2 – 4, positions where the vibration rate did not align with the stimulus frequency were selected for p1 and p8. If positions where the vibrato rate aligns with the stimulus frequency are used, the vibrato rate can be better controlled.
4.1.2 Experiment 2: current intensity
The results of Experiment 2 demonstrate that the vibrato extent increases with the current intensity (Figure 7A). This finding suggests that the vibrato extent can be controlled by adjusting the current intensity. As the current intensity increases, muscle activity appears to intensify, leading to greater changes in vocal cord length; thus, the range of F0 fluctuations increases. As shown in Figure 7B, the median vibrato rate matched the stimulus frequency (6 Hz) for some participants. The median vibrato rates for p4, p6, and p9 matched the stimulus frequency at 4 mA, whereas p2, p3, and p5 corresponded to 6 mA, and p1 and p7 corresponded to 10 mA. A low current intensity produces a small fluctuation in F0, making the voice more susceptible to the effects of natural fluctuations. In contrast, a high current intensity generates a larger vibrato extent, leading to a more stable F0 fluctuation. Therefore, when the current intensity increased, the vibrato rate tended to match the stimulation frequency.
4.1.3 Experiment 3: stimuli frequency
The results of Experiment 3 show that the median vibrato rate matched the stimulus frequency for 8 out of 11 participants (Figure 8B), indicating that the vibrato rate can be effectively controlled by the stimulus frequency. The fluctuation of F0 was large enough to prevent it from being affected by the natural fluctuation of the voice, and there was sufficient time for the muscles to relax. However, if the current intensity is insufficient, the natural fluctuation of the voice becomes relatively large, and this fluctuation can be determined by the vibrato rate.
However, the median vibrato rate did not match the stimulus frequency in three of the 11 participants (p8, p10, and p11), except for the 4 Hz stimulation of p8. There are three possible explanations for why the system did not produce vibrato. First, the skin resistance may have been high owing to dryness or thick skin, causing the current to flow easily over the skin surface, thereby not sufficiently stimulating the muscles. Skin resistance was resolved by carefully wiping the neck. Second, the laryngeal muscles may have become overstrained because of anxiety caused by the electrical stimulation. This resulted in a decrease in the number of muscle fibers newly induced by the stimulation, making it harder to produce vibrato. Such anxiety may be alleviated by adaptation. Third, the electrode position may not have been optimal. As shown in Figure 6B, p8 could not control the vibrato rate because an upper position was selected, even though the area where the stimulus frequency and vibrato rate matched was located below the electrode. An appropriate position can be identified by improving the electrode-search algorithm.
Not all the data from the other participants matched the stimulus frequency and vibrato rate. Differences in posture and muscle activity during speech appeared to affect the vibrato rate. With changes in the direction of the face, tilt of the neck, or position of the chin, the relative positions of the muscles and electrodes also change. Therefore, even if the same position is stimulated on the surface, the muscle group is stimulated, and its contraction changes, leading to a decrease in vibrato extent. Decreasing the vibrato extent makes the natural fluctuation of the voice more dominant.
As shown in Figure 8A, the vibrato extent tended to decrease as the stimulus frequency increased. This suggests that the contraction caused by the subsequent stimulation occurred before the effects of the previous stimulation were eliminated. Figure 10 shows the F0 waveform when 4, 6, 8, and 10 Hz stimuli were applied to p11. When simulating waveforms of other frequencies from a 4 Hz waveform, the extent was reduced following a similar slope in the simulation (Figure 8C, i.e., p3-5, p9). In contrast, p2 and p6 exhibited peaks at 8 Hz. One possible explanation is that muscle reflex resonance likely occurs in the 5–7 Hz range based on the vibrato reflex resonance model of .
FIGURE 10
4.1.4 Experiment 4: F0 pitch
The correlation test results showed a significant correlation between F0 and the vibrato extent for p3, p6, and p9. The vibrato extent tended to increase as the F0 pitch increased in p3 and p6, whereas it tended to decrease in p9. The influence of the upward and downward movement of the larynx could change the optimal electrode position. This implies that the electrode position must be adjusted according to the F0 pitch. Note that F0 did not fluctuate in the expected octave in many participants, and the data only had 16 or fewer measurements for each participant. In this study, the experiment was conducted at pitches that were comfortable for each participant, which limited the direct comparability of samples across participants and did not capture the wider range of fundamental frequencies involved in actual singing. In the future, we will conduct experiments over a broader range of F0 s and evaluate the effectiveness of adjusting the electrodes according to the F0 pitch.
4.2 Limitations
4.2.1 Participants
In this experiment, there were 14 participants, but data analysis was conducted on 11 of them, and vibrato was observed in 8 of those participants. This result indicates the feasibility of the proposed method, but also suggests that its stability and generalizability remain limited.
Participants’ formal musical background (e.g., vocal training history) and pitch accuracy measurements (e.g., pre-experiment pitch-matching tasks) were not systematically documented. Future studies should record and control for this factor.
4.2.2 Voluntary adjustment
Because electrical stimulation conveys both frequency and intensity to participants, voluntary adjustment cannot be completely ruled out. However, several aspects of our design mitigate this concern: (i) participants were not informed of the stimulation frequencies in advance; (ii) low current levels did not elicit sufficient vibrato; and (iii) participants were not instructed to generate vibrato. In addition, baseline fluctuations were confined to 3–4.67 Hz in all participants except p11, who exhibited broader individual variability, whereas stimulation-induced vibrato spanned a broader and controllable range, suggesting that the observed effects are more likely due to electrical stimulation than to inherent vocal variability. These fluctuation values are smaller than those observed under stimulation (median = 23.55 cents), supporting that the induced vibrato is not attributable to natural fluctuations alone. To more rigorously establish causality, future work will incorporate sham stimulation controls.
4.2.3 Current intensity and pain
Three out of the fourteen participants were unable to tolerate an increase in current intensity up to 10 mA (2.5, 3 and 6.5 mA, respectively). Because the throat is a sensitive area, electrical stimulation must be applied without pain. However, due to individual differences in pain perception and sensitivity, it may not be possible to provide completely painless stimulation for all subjects. Nevertheless, the following points may help reduce pain and broaden applicability. First, when the skin resistance is high, the current tends to concentrate on the skin surface. As a result, pain receptors on the skin surface become overstimulated, making it easier to feel pain. Second, anxiety related to electrical stimulation and anticipatory fear of pain may have contributed. Participants may have focused more strongly on pain due to the requirement to actively report it. Third, the electrodes exhibited low adherence to the skin. The participants could avoid pain by properly attaching the electrodes. Fourth, square-wave stimulation may induce discomfort owing to the rapid rise in current amplitude. This pain can be alleviated using a waveform that does not suddenly increase the current value. In addition, formal stress measures were not collected. Future work will incorporate quantitative stress assessment to more systematically characterize the psychological state of participants during stimulation.
4.2.4 Device limitations
The electrodes measured mm with a pitch of 2 mm. The stimulation area was mm, consisting of electrodes. However, alternative electrode sizes or placements may be more suitable for inducing vibrato. Smaller electrodes produce higher current densities, which may result in higher vibrato. As shown in Figure 6A, the maximum vibrato extent was located at the edges of the electrodes for p1, p3, p5, p7, p10, and p11. This suggests that stimulation sites outside the current electrode range may yield greater vibrato extent. The maximum current intensity was set to 10 mA to ensure safety. However, if the current were increased above 10 mA, the vibration rate could be controlled at p8, p10, and p11. In addition, a higher vibrato extent might have been achieved for the other participants. However, the higher the current, the more painful it becomes, inhibiting vibrato.
4.2.5 Potentially activated muscle
The muscles potentially activated in this study can be grouped into three categories based on their anatomical location and likely contribution to the observed vocal effects. First, the platysma may have caused the visible trembling of the throat and jaw seen in some participants; although such extra-laryngeal movements do not directly affect vocal fold vibration, they can indirectly alter F0 via changes in vocal tract geometry and laryngeal position. Second, suprahyoid and infrahyoid muscles may have modulated F0 through hyoid displacement or thyroid cartilage. Third, direct activation of intrinsic muscles such as the thyroarytenoid and cricothyroid cannot be ruled out, though their deeper location may reduce the likelihood of direct activation by surface stimulation. Future work will employ EMG, ultrasound imaging, and/or electric field simulation to identify which muscles were actually activated.
4.3 Future work
4.3.1 Wearable devices
In this study, stationary microphones were used to record high-precision data. In addition, the stimulus parameters were controlled via communication with a PC. In the future, a wearable vibrato-control system for everyday use may be developed using a small microphone and controlling stimulus parameters with a microcomputer.
4.3.2 Vibrato for songs
The sound produced in the experiment was the vowel “/a/.” Future studies will examine whether vibrato can be controlled via electrical stimulation for other vowels, such as “/i/” or “/u/,” when singing a song. In the future, we aim to develop a system that can apply vibrato at any desired depth and rate using electrical stimulation. Such a system could enable the automated application of vibrato by selecting the section of the song where the vibrato is to be applied, as well as specifying the desired extent and rate. In addition, by analyzing a singer’s vibrato in advance, it would be possible to replicate their vibrato parameters, enabling the singer to perform with consistent vibrato.
4.3.3 Vibrato training
The distinctive feature of our approach is that it directly modulates the user’s own voice in real time, in contrast to post-processing applied to a recorded voice. This enables learners to experience the physical sensation and auditory feedback of vibrato applied to their own vocalization, which may serve as a supportive tool in vocal pedagogy. Whether such exposure produces lasting improvements in voluntary vibrato control, and how it interacts with traditional pedagogical methods, remains an important direction for future research. More broadly, the ability to directly modulate the singer’s own voice in real time offers a new mode of vocal expression control, which may have potential applications in the field of singing.
4.3.4 Vibrato jockey
By applying this system, external performers can control the singer’s vibrato parameters. Specifically, the vibrato rate is adjusted according to the stimulus frequency, and the vibrato extent is controlled by the external current intensity. When the performer adjusts the rate and extent of vibrato on a mobile device, the vibrato characteristics of the singer’s voice change. This system harmonizes the system output with the user’s singing expression, creating an interactive experience in which the live singing voice becomes an important part of the musical performance. This experience is similar to that of a DJ manipulating sound.
4.3.5 The listeners impressions
This paper described a device that enables control of vibrato extent and rate through electrical stimulation. However, the perceptual impressions of listeners regarding this modulated vibrato remain unclear. Previous research has suggested that electrical stimulation of the larynx can evoke sensations of changes in vibrato extent and rate based on user studies (). To further advance this line of work, it will be necessary to conduct subjective evaluations to determine which frequencies and intensities are perceived as beautiful or as natural vibrato.
4.3.6 Immersion in virtual reality avatars
While our results demonstrate the feasibility of controlling vocal vibrato through selective electrical stimulation, the application of this technique to VR systems remains a promising but untested direction. Because vocal production is closely tied to embodiment and self-perception, extending our method to VR could potentially enrich avatar interaction and enhance immersion. Vibration affects the state of embodiment toward virtual reality avatars (; ). In this context, vocal modulation at the motor level may provide a technical basis that connects to embodied interactions in virtual reality, particularly those involving virtual reality avatars. For example, controlling vocal vibrato corresponding to a character designed with a trembling or vibrato voice can be explored as a means of aligning vocal expression with avatar characteristics, potentially contributing to immersive virtual experiences. However, verifying this would require dedicated VR scenario validation, including avatar interaction tasks and user-reported immersion evaluations, which we plan to address in future work.
5 Conclusion
This study aimed to develop selective electrical stimulation of the larynx to control vibrato. The following four experiments were conducted: In Experiment 1, we found that the vibrato extent changed depending on the position of the stimuli. Experiment 2 revealed significant differences in the vibrato extent. In Experiment 3, we discovered that the median vibrato rate matched the stimulus frequency for eight out of 11 participants. In addition, Experiment 4 showed a significant correlation between F0 and the vibrato extent in three participants. These results suggest that vibrato can be controlled by electrical stimulation of the larynx by controlling the stimulus parameters. This technology is expected to enable control of vibrato in human vocal cords.
In the future, we aim to apply this system to real singing, allowing singers to freely adjust the vibrato extent and rate according to the song, thereby enhancing their singing ability. For instance, singers can emphasize vibrato during specific phrases or increase its extent at the climax of a song. This advancement will not only enhance a singer’s expressive capabilities but also enrich the listening experience for the audience.
Furthermore, vibrato modulation through electrical stimulation alters a person’s own motor actions, which is important for interactions with the environment, as discussed in virtual reality. Thus, vibrato-control technology using laryngeal electrical stimulation is the first step toward more fine grained control of vocal embodiment.
Statements
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Ethics statement
The studies involving humans were approved by All experiments were approved by the Life Science Research Ethics and Safety Office of the University of Tokyo, Japan (approval number: 23–350). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.
Author contributions
DU: Formal Analysis, Data curation, Investigation, Writing – original draft, Conceptualization, Software, Methodology. SM: Project administration, Conceptualization, Formal Analysis, Software, Methodology, Visualization, Validation, Writing – original draft, Funding acquisition, Writing – review and editing, Resources. AM: Supervision, Validation, Writing – review and editing, Conceptualization.
Funding
The author(s) declared that financial support was received for this work and/or its publication. This work was supported by JSPS KAKENHI Grant Number JP24K20825.
Acknowledgments
We would like to thank Alvaro Costa Garcia for his assistance with the preparation of this manuscript. We would like to thank Editage (www.editage.jp) for English language editing.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1
AnandS.WingateJ. M.SmithB.ShrivastavR. (2012). Acoustic parameters critical for an appropriate vibrato. J. Voice26 (6), 820.e19–820.e25. 10.1016/j.jvoice.2012.06.004
2
AndradeP. A.FričM.OtčenášekZ. (2023). Assessment of changes in laryngeal configuration and voice parameters among different frequencies of neuromuscular electrical stimulation (nmes) and cumulative effects of nmes in a normophonic subject: a pilot study. J. Voice37 (4), 632.e1–632.e20. 10.1016/j.jvoice.2021.03.018
3
BanakouD.SlaterM. (2014). Body ownership causes illusory self-attribution of speaking and influences subsequent real speaking. Proc. Natl. Acad. Sci.111, 17678–17683. 10.1073/pnas.1414936111
4
BanakouD.SlaterM. (2017). Embodiment in a virtual body that speaks produces agency over the speaking but does not necessarily influence subsequent real speaking. Sci. Reports7, 14227. 10.1038/s41598-017-14620-5
5
CarterN. (2010). Volitional control of vibrato in trained singers. J. Singing-The Official J. Natl. Assoc. Teach. Sing.67, 9–17.
6
DromeyC.CarterN.HopkinA. (2003). Vibrato rate adjustment. J. Voice17, 168–178. 10.1016/S0892-1997(03)00039-0
7
EkholmE.PapagiannisG. C.ChagnonF. P. (1998). Relating objective measurements to expert evaluation ofvoice quality in western classical singing: critical perceptual parameters. J. Voice12, 182–196. 10.1016/S0892-1997(98)80038-6
8
EricksonM. L.GaskillC. S. (2012). Can listeners hear how many singers are singing? The effect of listener’s experience, vibrato, onset, and formant frequency on the perception of number of simultaneous singers. J. Voice26 (6), 817.e1–817.e13. 10.1016/j.jvoice.2012.04.011
9
FerranteI. (2011). Vibrato rate and extent in soprano voice: a survey on one century of singing. J. Acoust. Soc. Am.130, 1683–1688. 10.1121/1.3621017
10
FowlerL. P.Gorham-RowanM.HapnerE. R. (2011). An exploratory study of voice change associated with healthy speakers after transcutaneous electrical stimulation to laryngeal muscles. J. Voice25, 54–61. 10.1016/j.jvoice.2009.07.006
11
FushimiR.FujinawaE.NarumiT.TanikawaT.HiroseM. (2017). “Vibrat-o-matic: producing vocal vibrato using ems,” in Proceedings of the 8th Augmented Human International Conference (New York, NY, USA: Association for Computing Machinery), AH ’17), 1–5. 10.1145/3041164.3041193
12
Gorham-RowanM.FowlerL.HapnerE. (2010). Acoustic analysis of voice change in normal speakers following transcutaneous electrical stimulation to the laryngeal area. Open Rehabil. J.3, 67–74. 10.2174/1874943701003010067
13
InbarG. F.EdenG. (1983). Physiological evidence for central modulation of voice tremor. Biol. Cybern.47, 1–12. 10.1007/BF00340063
14
JaschkeA. C.EggermontL. H.HoningH.ScherderE. J. (2013). Music education and its effect on intellectual abilities in children: a systematic review. Rev. Neurosci.24, 665–675. 10.1515/revneuro-2013-0023
15
KamiokaH.TsutaniK.YamadaM.ParkH.OkuizumiH.TsuruokaK.et al (2014). Effectiveness of music therapy: a summary of systematic reviews based on randomized controlled trials of music interventions. Patient Preference Adherence8, 727–754. 10.2147/PPA.S61340
16
KempsterG. B.LarsonC. R.KistlerM. K. (1988). Effects of electrical stimulation of cricothyroid and thyroarytenoid muscles on voice fundamental frequency. J. Voice2, 221–229. 10.1016/S0892-1997(88)80080-8
17
KingJ. B.HoriiY. (1993). Vocal matching of frequency modulation in synthesized vowels. J. Voice7, 151–159. The Voice Foundation’s 22nd Annual Symposium. 10.1016/S0892-1997(05)80345-5
18
KondoR.SugimotoM.MinamizawaK.HoshiT.InamiM.KitazakiM. (2018). Illusory body ownership of an invisible body interpolated between virtual hands and feet via visual-motor synchronicity. Sci. Reports8, 7541. 10.1038/s41598-018-25951-2
19
LopesP.YouS.ChengL.-P.MarweckiS.BaudischP. (2017). Providing Haptics to Walls and Heavy Objects in Virtual Reality by Means of Electrical Muscle Stimulation in Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (New York, NY, USA: Association for Computing Machinery), 1471–1482. 10.1145/3025453.3025600
20
MatsubaraS.WatanabeT.SuzukiT.WakisakaS.AoyamaK.InamiM. (2023). Optimal motor point search using mm-order electrode arrays. IEEE Access11, 58970–58981. 10.1109/ACCESS.2023.3285422
21
MitchellH. F.KennyD. T. (2010). Change in vibrato rate and extent during tertiary training in classical singing students. J. Voice24, 427–434. 10.1016/j.jvoice.2008.12.003
22
MoriseM. (2017). Harvest: a high-performance fundamental frequency estimator from speech signals. Interspeech2017, 2321–2325. 10.21437/Interspeech.2017-68
23
MürbeD.ZahnertT.KuhlischE.SundbergJ. (2007). Effects of professional singing education on vocal vibrato—a longitudinal study. J. Voice21, 683–688. 10.1016/j.jvoice.2006.06.002
24
NakanoT.GotoM.HiragaY. (2006). “An automatic singing skill evaluation method for unknown melodies using pitch interval accuracy and vibrato features,” in Interspeech 2006. paper 1854–Wed2A1O.4. 10.21437/Interspeech.2006-474
25
NixJ. P. (2015). Listener preferences for vibrato rate and extent in synthesized vocal samples. J. Acoust. Soc. Am.137, 2404–2405. 10.1121/1.4920757
26
PrameE. (1994). Measurements of the vibrato rate of ten singers. J. Acoust. Soc. Am.96, 1979–1984. 10.1121/1.410141
27
PrameE. (1997). Vibrato extent and intonation in professional western lyric singing. J. Acoust. Soc. Am.102, 616–621. 10.1121/1.419735
28
RaikovaR. T.AladjovH. T. (2002). Hierarchical genetic algorithm versus static optimization—investigation of elbow flexion and extension movements. J. Biomechanics35, 1123–1135. 10.1016/S0021-9290(02)00031-3
29
RaikovaR.CelichowskiJ.PogrzebnaM.AladjovH.KrutkiP. (2007). Modeling of summation of individual twitches into unfused tetanus for various types of rat motor units. J. Electromyogr. Kinesiol.17, 121–130. 10.1016/j.jelekin.2006.01.005
30
SaitouT.GotoM. (2009). “Acoustic and perceptual effects of vocal training in amateur male singing,” in Interspeech 2009, 832–835. 10.21437/Interspeech.2009-65
31
SeashoreC. E. (1931). The natural history of the vibrato. Proc. Natl. Acad. Sci.17, 623–626. 10.1073/pnas.17.12.623
32
SundbergJ. (1995). Acoustic and psychoacoustic aspects of vocal vibrato. Vibrato35, 35–62.
33
TitzeI. R.StoryB.SmithM.LongR. (2002). A reflex resonance model of vocal vibrato. J. Acoust. Soc. Am.111, 2272–2282. 10.1121/1.1434945
Summary
Keywords
electrical stimulation, human augmentation, human computer interaction, human voice, vibrato
Citation
Uchida D, Matsubara S and Murai A (2026) Controlling vocal vibrato: a feasibility study of vibrato modulation using selective electrical stimulation of the larynx. Front. Virtual Real. 7:1869451. doi: 10.3389/frvir.2026.1869451
Received
30 April 2026
Revised
09 July 2026
Accepted
28 July 2026
Published
07 September 2026
Volume
7 - 2026
Edited by
Sunyoung Jang, Suny Upstate Medical University, United States
Updates
Copyright
© 2026 Uchida, Matsubara and Murai.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Seito Matsubara, seito-matsubara@aist.go.jp
† These authors have contributed equally to this work and share first authorship
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.