Emotion expression in the singing voice: Testing a parameter modulation technique for improving communication of emotions through voice qualities

This study examines emotional expression in singing and its teachability using a novel parameter modulation technique. The work is an experimental comparative study using listener evaluations, acoustic analyses, and statistical deduction to assess the emotional expressiveness of the singing voice from short vocal samples and sung phrases. The investigation consists of three sub-studies, the first of which explores the auditory recognition of emotion from samples sung with Classical and non-Classical singing techniques at three different pitches. The second study compares the qualitative features of emotional expression in Classical and non-Classical singing techniques by means of acoustic analysis. The third sub-study focuses on teaching the parameter modulation technique to acting students. It compares the clarity of emotional expression between the instructional and control groups before and after the training intervention. The measures of emotional expression clarity in this study are considered to be the auditory recognition of emotional expression and the qualitative variation of the voice between different emotional expressions.

The study involved 29 (Study I) and 32 (Study III) listeners of sound samples, 11 female singers (six with Classical singing technique training and five with popular music singing technique training) (Studies I & II), two male singers (one with Classical singing technique training and one with popular singing technique training) (Study I), and six + six acting students who gave song samples, one group of whom participated in the parameter modulation training while the other group received standard singing training (Study III). Listeners were to classify samples into neutral expressions and expressions of joy, tenderness, sadness, and anger from short vowel samples and phrases. The emotions were chosen because of their opposite positioning on the valence-activation scale. Singers sang spontaneous emotional expression into short melodies (16 bars in Studies I & II & 8 bars in Study III) from which sound samples were cut. Samples of the sung [a:] vowel were analyzed with the Praat sound analysis program. The samples were analyzed for fundamental frequency (fo), sound pressure level (SPL), formant frequencies (F1-F5), harmonics- to-noise ratio (HNR), energy ratio between upper and lower frequencies of the spectrum (Alpha ratio), irregular cycle-to-cycle variation of fundamental frequency (Jitter rap & ppq5), irregular cycle-to-cycle variation of amplitude (apq3 & apq5), vibratos (fo rate and extent & rate and extent of amplitude), and amplitude contour: attack, sustain, release.

The results of the study showed that emotional expression can be identified from the singing voice when singers express emotion spontaneously (Studies I & III). In this study, the identification of emotional expression became easier after singers received instructions on the use of the parameter modulation technique (Study III). Emotional expression was better identified from song samples sung in a non-Classical style (Study I). Pitch, SPL, emotional valence (positive / negative), and activation level (high / low) had an effect on emotional recognition (Studies I-III). SPL, Alpha ratio, and HNR values increased in expressions of high activity emotions (anger and joy) and decreased in expressions of low activity emotions (sadness and tenderness), suggesting increased muscle activity and tighter vocal fold adduction in high energy emotions (Studies II & III). Formants packed in high-energy emotions and scattered in low-energy emotions, suggesting a modification of the vocal tract for the expression of different emotions (Studies II & III). Jitter and shimmer were more prevalent in low-energy emotions, suggesting lower muscle activity (Study II). Fundamental frequency vibrato was slower in Classically trained singers (Study II), whereas in non-Classical singers, amplitude vibrato was statistically significant in differentiating emotions (Study II). Vocal offsets were statistically significant in terms of emotional expression in singers singing with a non-Classical singing technique (Study II).

The main question of the study was whether it is possible to integrate vocological research data on the acoustic parameters of emotional expression into practical singing exercises and thus enhance emotional expression in the singing voice. In the study, we used a seven-week training program focusing on parameter modulation techniques that taught the use of different sound qualities to a group of acting students. A similar group of acting students who did not receive special training served as a control group. The test group increased the use of different sound qualities as a means of emotional expression after training. This result was confirmed by acoustic analyses and improved recognition of emotions by the listeners. The control group did not show such an effect. After training, the test team appeared to use F1, SPL, HNR, and alpha ratio for emotional expression more systematically.

The study showed that the sound pressure level and the way energy is distributed in the sound spectrum were the two most typical indicators of the emotional characteristics of sound. The study finds that training in different sound qualities can help with the expression of emotions in the singing voice.