Contact Information
last updated: 10/22/2010

Learning Timbre


     One of the objectives of the Indiana University International Vocal Ensemble (IVE) is to sing music of non-western cultures in the native language, and to the degree possible, sing with integrity of vocal and musical style. A subsequent objective is to imitate music accurately when presented with an aural model. With these objectives in mind I am interested in semantic perception and computer analysis cues of vocal timbre and the degree, if any, to which vocal timbre can be imitated between cultures, and how this information can be added to a transcription to aid the singer. For this particular project I focused on the production of Ghanaian vocal timbre by three native Ghanaians, individuals in IVE, and non-IVE members of a Ghanaian children's song. This paper will focus on the data of one of the Ghanaian female informants (Gf1) and one American female IVE member (Af1).


     There is no conclusive theory of timbre perception. Part of the difficulty is that timbre is difficult to perceive as an isolated phenomenon since it is based on perception. Timbre is connected with the source as opposed to individual and measurable attributes such as frequency or amplitude. With regards to vocal timbre the vocal tract transforms the airflow spectrum into recognizable acoustical patterns which we know as vowels. Vowels are a musical element of singing aside from their information carrying function. (Benade, 1990) Three areas of vowel production that are typically studied in computer analysis when researching timbre are the harmonic spectra, formants, and attack and decay transients.

     Another approach to investigating timbre along with computer analysis is the investigation of the verbal descriptions of timbre provided by subjects. Such descriptions yield a "semantic scale" in regard to specific tone qualities. In their 1988 study, Bloodcroft and Plomp collected fifty scales from related studies on timbre and literature from singing, and created a list of twenty-one pairs suitable as common adjectives used as a description of singing vocal timbre including light/dark, open/covered, metallic/velvety, angular/round, clear/dull, open/throaty, rough/smooth, etc.

I was interested in both acoustical properties and semantic descriptions of vocal singing timbre. The results from this project can in no way make large scale evaluations of Ghanaian and American timbre, but should be regarded as a stepping stone into the investigation of timbre perception and imitation among Ghanaian and American voices.


     I began by recording three native Ghanaian speakers singing one of the songs that was currently being rehearsed in IVE, the Ghanaian children's song "Kofi ". (Appendix 1) Following the recording, and an aural example of an American singing the song I asked the subsequent interview questions 1) Do you hear a difference in the tone color of Western vocalists as compared to your own? 2) How would you describe the Ghanaian timbre as compared to American timbre? 3) What do you physically feel when you are singing this children's song?

The IVE members (solo) sang the song twice. The first time they sang without any instruction as to vocal production, and once again after hearing the tape of one of the Ghanaian singers they were asked to try to imitate the timbre of the voice they heard. I asked the Americans the same questions as the Ghanaians, with the added question of what did you do differently, either physically or mentally when you tried to imitate Ghanaian timbre?


     Using the Sound and Visual Analysis Instructional Lab (SAVAIL) at the Indiana Folklore Institute, and previous research of vocal timbre, the first comparison in the lab was between informant Gf1 and informant Af1 singing the song in their own timbre. Both female singers had complementary formant patterns with high harmonics. The harmonics in informant Af1's spectrogram were much clearer, whereas informant Gf1's seemed to be surrounded by noise. (figure 1) I applied a filtering process to both voices for a three syllable phrase. Beginning by filtering out high frequency noise, I eventually filtered the samples to a point where there was low noise presence in both samples, and only slight differences between the third and fourth harmonic in both samples. Even with all acoustic information above, and noise above about 1000 Hz filtered out, timbral difference was still audible.

     The unfiltered spectrogram of informant Gf1 revealed several consonants to have been articulated in an explosive manner, whereas informant Af1's consonant attacks in general were not as prominent. These dominant consonant attacks were evident both perceptually on the recordings and visually in the computer displays. (figure 2)

     Some of the verbal choices used by the IVE members to describe Ghanaian voices were verbal descriptions of a problem, eg. nodes on the voice, clogged throat, etc. Further investigation might pursue this topic with questions like "What do you regard as a "good" voice?". When comparing informant Af1's normal singing with her imitation of the Ghanaian singing, while there was an audible difference between the two, a seemingly softer tone in her imitated example, the spectral analysis showed only a slight weakening of her upper harmonics. Informant Af1's instincts were to soften the consonants, whereas the computer showed evidence that informant Gf1 did the opposite.

Various methods regarding how the singer can alter vocal timbre have been researched. In a 1970 study by Sundberg, acoustic and x-ray evidence was cited that shows that Western trained singers utilize a low larynx position throughout their vocal range resulting in a strong resonance in the spectrum in an area he described as the "singer's formant" (fourth formant). In addition to larynx height, tongue position, and lip and jaw movement may also contribute to acoustical features. (Sundberg, 1987)

     Interestingly enough, it was the formants which were most alike in the Ghanaian and American voices, so Sundberg's findings are only partially relevant. A closer investigation of the formant frequencies of a larger number of Ghanaian and American informants would be necessary for a more comprehensive study on timbral differences in these examples.

It is important to note that most people have a low awareness of articulatory activity, and therefore control of articulatory devices is not easily possible. In the case of teaching a choir to imitate a variety of different vocal timbres, this type of time spent focusing on variant formant frequencies of a number of different vocalists becomes unrealistic.


     In Charles Seeger's article "Prescriptive and Descriptive Music Writing" he describes a transcription as prescriptive in character if it has emphasis on structures - pitch and meter, but not the connection of the two. He goes on to say the 1)we ignore what has no symbols and 2) we expect the non-tradition bearers to read it. I have attempted to transcribe "what happens between the notes" and how this relates to timbre. I have chosen to focus on the attack and decay transients and their placement as displayed in informant Gf1's consonants as well as semantic descriptions.

     When focusing on timbre in the transcription I wanted to have a few general descriptive words at the beginning of the piece to orient the singer to the type of tone and style of the piece, followed by visually stimulating evidence of how to approach the consonants. (Appendix 2) The symbols correspond to the shape of the consonant attack or decay from the computer spectrogram. (Figure 3, 4, 5) I brought the transcription to two non IVE vocalists to see their perceptions and reactions to the piece and the devised notation of timbre. The shapes in the transcription were an obvious cue for the various consonant distinctions, and it was during those syllables of the song that they most sounded like the Ghanaian example that I knew.


     It now appears that both a combination of the auditory stimulus and written transcription would prove most beneficial for a group such as IVE in imitating timbre. Along with an understanding of the components involved in creating timbre, an American singer most benefits when first introduced to a song aurally, and then uses a description, or score for visual cues and reminders of how to alter timbre. Rather than specializing on a long term basis in the repertoire of one area, the majority of groups performing non-Western music choose repertoire from various areas, in the attempt to imitate various timbres. What is important is that the instructor be clear and knowledgeable as to the qualities of timbre, and decide to what extent he/she would like to focus on timbre with the choir, and which models to follow.

     While this study focused on individual voices, the context of the timbre of the song in a choral performance is a further avenue of study. Most research on singing has been concerned with the solo operatic voice. A 1980 study by Goodwin compared how sopranos used their voices in a choral situation as opposed to singing solo. Rossing et al. (1985 and 1986) studied both male and female singers, comparing use of the voice as a choral singer and as a soloist. The results from both studies indicate that use of the voice differs in choral singing and solo singing, and in Western trained singing a different timbre is desired for choral singing compared with solo singing. Another difference is that "...choral singers strive to tune their voice timbre in order to mesh with the timbre of the rest of the choir, while a solo singer would try to develop his or her own individual timbre." (Sundberg, 1987) How do people's perceptions of timbre translate from the individual to a group sound? What adjustments do one's ears make? These issues are open for further research on the timbre of singing.

     The song Kofi has since been performed at numerous concerts, and many people in the choir will continue humming the song when it pops into their head for one reason or another. Next semester there will be new pieces and new timbres to begin to model, with new research ready to be investigated.

last updated: 10/22/2010