467 | 2013 ICMC idea | DEMONSTRATIONS A TYPE-BASED APPROACH TO GENERATIVE PARAMETER MAPPING Josu´ e Moreno Independent Researcher & Composer Valencia, Spain Vesa Norilo Sibelius Academy Centre for Music & Technology Helsinki, Finland ABSTRACT This paper presents a type-based strategy for automat- ing parameter mapping. It approaches the problem of in- tegrating a disparate set of control signals into a musical application while minimizing the amount of boilerplate code the user has to write in order to connect control sig- nals to unit generators. The method is based on polymorphic mapper compo- nents and signal metadata that allows typed connections. Concepts for a semantically meaningful parameter map- ping system for musical applications are explored and im- plemented in the Kronos programming language. 1. INTRODUCTION Much of digital instrument building is point-to-point rout- ing of signals. Often, these signals will need to be adapted on the go: for an instrument to be playable, its control sur- face must be mapped so that most of its parameter space is musically meaningful, maximizing expressiveness while being as economical as possible. This results in carefully arranged adaptation layers, where the output from a control is scaled, translated and transformed to a range that is appropriate for an audio generator. Changing either the control or the generator typically means that the entire adaptation layer must be redesigned. Little work has been done to standardize the funda- mental mapping algorithms [6]. Likewise, few approaches have been made towards an automatization of mapping, with the exception of a neural networks based gesture iden- tification and mapping libraries [8], which are not a gen- eral purpose solution. With no standard framework, instrument designers are constantly re-implementing mapping algorithms whenever creating a new instrument. Many mapping operations are very common, so it makes sense to design a general pur- pose software library that includes these operations [6]. Not all the composers or instrument designers have the necessary background to envision proper mapping schemes, resulting in linear and oversimplified parameter transla- tion. These are rarely musically satisfying. The authors propose a method to perform automatic parameter mapping based on an adaptation layer that can automatically reconfigure itself based on the semantics of both its input and output. This paper is organized as follows; in Section 2, Previ- ous Work, the current state of parameter mapping methods is examined. In Section 3, Type-based mapping, the the principles of such mapping strategy will be stated and im- plementation details discussed. Finally, in Section 4, Ex- amples, the authors will demonstrate the features of type- based mapping. 2. PARAMETER MAPPING STRATEGIES 2.1. Fundamentals of Data Mapping Data mapping is the process of creating data element map- pings between two distinct data models. It consists of data transformation or data mediation between a data source and a destination. Data-driven mapping is a recent approach in data map- ping and involves simultaneously evaluating actual data values in two data sources using heuristics and statistics to automatically discover complex mappings between two data sets. Semantic Mapping consists of, given a data set, con- structing a projection matrix that can be used for mapping data elements from one dimensional space into another, being or not of the same dimension count. It is an alter- native to random mapping, principal components analysis and latent semantic indexing methods. [4] Mapping in musical applications, has been mainly fo- cused on several strategies formalized as: One to one; One to many; Many to one; Many to many. These have been implemented by the composers themselves or within a software application [10] that may carry aesthetic im- plications. Also, this categorization is not so much a map- ping strategy as an overrall observation about different in- put/output configurations one may encounter. 2.2. Survey of the Field Previously, Steiner has made work towards a Catalog of Mapping algorithms for the purpose of developing a Map- ping Framework within Pd. [6] Continuing on previ- ous developments such as the [hid] library [7], MnM for Max/MSP [2] and ESCHER [5]. In these frameworks interaction between mapping modules is based on stan- dardizing all parameter ranges to a linear range of [0, 1]. Any information concerning the range properties (multi- dimensional, logarithmic or bipolar) is dropped. Another approach using normed parameters, in this case of [-1, 1],