Computational auditory scene analysis (CASA) provides an excellent means to improve speech intelligibility in adverse acoustical situations. In order to utilize algorithms of CASA in hearing aids, sets of algorithmic parameters need to be adjusted to the individual auditory performance of the listener and the acoustic scene in which they are employed. Performed manually, the optimization is an expensive procedure. We therefore developed a framework in which algorithms of CASA are automatically optimized by the principles of evolution, i.e., by a genetic algorithm. By using the speech transmission index (STI) as an objective function, the presented framework presents a holistic routine which is solely based on psychoacoustical and physiological models to improve and to assess speech intelligibiltiy. The initial listening test revealed a discrepancy between the objective and subjective assessement of speech intelligibility, which suggests a review of the objective function. Once the objective function is in accordance with the individual perception of speech intelligibility, the presented framework could be applied in the optimization of all complex speech processors and therewith accelerate their assessment and application.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.