Modelling binaural speech intelligibility against a harmonic masker
Speech intelligibility models are able to predict intelligibility of a target voice in the presence of non-stationary noise interferers. However, there is currently no model able to accurately predict intelligibility in the presence of competing speech maskers. Our aim is to develop a binaural intelligibility model able to do so. As a first step, we need to accurately predict energetic masking for speech maskers, which will then allow us to quantify informational masking. Contrary to a noise signal, a speech signal has a harmonic structure that allows for F0 segregation. F0 segregation can be due to either spectral glimpsing or harmonic cancellation, but it is unclear what the relative contributions of these two mechanisms are to F0-based release from masking. In this work we have modified the model of Collin and Lavandier (2013) in order to take into account spectral glimpsing, and we are currently working on including harmonic cancellation. The model is being applied to two data sets. Leclère et al. (2017) measured SRTs for harmonic maskers that varied in their fundamental frequencies, temporal envelope and spatial position. Deroche et al. (2014) also measured SRTs for harmonic maskers that varied in their fundamental frequencies and degree of harmonicity. These two data sets represent a step between noise and speech masking as they include F0 differences but no informational masking. Comparison of model predictions to the two data sets will allow us to establish to what extent the model can predict F0 segregation for harmonic maskers, and determine the relative roles of spectral glimpsing and harmonic cancellation.