My research at the London College of Communication is concerned with archives of recorded speech, what new tools need to be devised for its manipulation and how to go about this process of invention. Research into available forms of analysis of speech is discussed below with regard to two specific areas, feature vectors from linear predictive coding (LPC) analysis and hidden Markov-model-based automatic speech recognition (ASR) systems. These are discussed in order to demonstrate that whilst aspects of each may be useful in devising a system of speech-archive manipulation for artistic use. Their drawbacks and deficiencies for use in art – consequent of the reasons for their invention – necessitate the creation of tools with artistic, rather than engineering agendas in mind.

It is through the initial process of devising conceptual tools for understanding speech as sound objects that I have been confronted with issues of semiotics and semantics of the voice and of the relationship between sound and meaning in speech, and of the role of analysis in mediating existing methods of communication. This is discussed with reference to Jean-Jacques Nattiez's Music and Discourse: Towards a Semiology of Music (Nattiez 1987). The ‘trace’ – a neutral level of semiotic analysis proposed by Nattiez, far from being hypothetical as suggested by Hatten (1992: 88–98) and others, is present by analogy to many forms of mediation in modern spoken communication and the reproduction of music, and it is precisely this neutrality with regards to meaning that tools for manipulation of speech must possess, since the relationships between the sound of speech and its meaning are ‘intense’ (after Deleuze 1968).