Technical description

Many different auditory models are available that can transform an input signal into an auditory representation. The actual design challenges behind the Auditory front-end arise from the multiplicity of supported representations, the requirement to process continuous signal in a chunk-based manner, and the ability to change what is being computed at run-time, which will allow the incorporation of feedback from higher processing stages. In addition to these three constraints, the framework will be subject to frequent updates in the future of the Two!Ears project (e.g., adding new processors), so the expandability and maintainability of its implementation should be optimal. For these reasons, the framework is implemented using a modular object-oriented approach.

This chapter exposes the architecture and interactions of all the objects involved in the Auditory front-end and how the main constraints were tackled conceptually. In an effort to respect encapsulation and the hierarchical organisation of the objects, the sections are arranged in a “bottom-up” way: from the most fundamental objects to the more global processes.

All classes involved in the Auditory front-end implementation are inheriting the Matlab handle master class. This allows every created object to be of the handle type, and simulates a “call-by-reference” when manipulating the objects. Given an object obj inheriting the handle class, doing obj2 = obj will not copy the object, but only obtain a pointer to it. If obj is modified, then so is obj2. This avoids unnecessary copies of objects, limiting memory use, as well as providing user friendly handles to objects included under many levels of class hierarchy. The user can manipulate a simple short-named handle instead of tediously accessing the object.