Previous neuroimaging studies have shown an increased sensory cortical response (i.e., heightened weight on sensory evidence) under higher levels of predictive uncertainty. The signal enhancement theory proposes that attention improves the quality of the stimulus representation, and therefore reduces uncertainty by increasing the gain of the sensory signal. The present study employed functional magnetic resonance imaging (fMRI) to investigate the neural correlates for ambiguous valence inferences signaled by auditory information within an emotion recognition paradigm. Participants categorized sound stimuli of three distinct levels of consonance/dissonance controlled by interval content. Separate behavioural and neuroscientific experiments were conducted. Behavioural results revealed that, compared with the consonance condition (perfect fourths, fifths and octaves) and the strong dissonance condition (minor/major seconds and tritones), the intermediate dissonance condition (minor thirds) was the most ambiguous, least salient and more cognitively demanding category (slowest reaction times). The neuroscientific findings were consistent with a heightened weight on sensory evidence whilst participants were evaluating intermediate dissonances, which was reflected in an increased neural response of the right Heschl's gyrus. The results support previous studies that have observed enhanced precision of sensory evidence whilst participants attempted to represent and respond to higher degrees of uncertainty, and converge with evidence showing preferential processing of complex spectral information in the right primary auditory cortex. These findings are discussed with respect to music-theoretical concepts and recent Bayesian models of perception, which have proposed that attention may heighten the weight of information coming from sensory channels to stimulate learning about unknown predictive relationships.