A two-dimensional space is proposed for exploration and interactive design in the sonic space of a sound model. A number of reference items, positioned as landmarks in the space, contain both a synthetic sound and its vocal imitation, and the space is geometrically arranged based on the acoustic features of these imitations. The designer may specify new points in the space either by geometric interpolation or by direct vocalization. In order to understand how the vast and complex space of the human voice could be organized in two dimensions, we collected a database of short excerpts of vocal imitations. By clustering the sound samples on a space whose dimensionality has been reduced to the two principal components, it has been experimentally checked how meaningful the resulting clusters are for humans. The procedure of dimensionality reduction and clustering is demonstrated in the case of imitations of engine sounds, giving access to the sonic space of a motor sound model.
INTRODUCTIONIn a sense, the human voice has for acoustic communication a role similar to what the hand and pencil have for visual communication. Humans use their voice for verbal communication as well as for non-verbal acoustic expression, similar to the hand which is used both for writing and for drawing. Just as the hand and pencil are extensively used for visual sketching, the voice has potential to be exploited for sketching or imitating sounds. Indeed, sketching comes before verbal-oral or written-expression in development of both the human species and the human individuals [4]. Recent research has shown that vocal imitations can be more effective than verbalizations at representing and communicating sounds [16]. Such natural capabilities are being exploited for sound retrieval and synthesis [3]. The European project SkAT-VG is investigating the use of voice and gesture as intuitive means for the selection and control of sound models in sonic interaction design [25].In this article we investigate how the vocal mimicking capabilities of humans may be exploited to access and explore a given sonic space. In this context, a sonic space is the space where the sounds produced by a given sound model can be distributed. Essentially, the multidimensional range of possibilities of a sound generator defines its sonic space.The driving idea of this work is that a collection of exemplar sounds produced by the sound model could be coupled to corresponding vocal imitations, thus providing landmarks to navigate the sonic space. The spatial organization of the exemplars should result in a low-dimensional layout where such landmarks can prompt exploration by vocal imitation or by geometric interpolation.The construction and use of a two-dimensional space is explained in Sec. 1, where the synthesis-analysisresynthesis loop involving human imitations is introduced. Since vocal imitations would be used to orient the user in a low-dimensional space of sounds, it is interesting to understand what is the span of the space of vocal imitations at large and to chec...