Patterns of learning and generalization seen after training on sinusoidal-amplitude-modulation (SAM) rate-discrimination and SAM-detection tasks

The perception of fluctuations in sound amplitude is crucial for speech understanding. Two psychophysical tasks that assess the ability of listeners to perceive amplitude modulation are sinusoidal-amplitude-modulation (SAM) detection and SAM rate discrimination. In SAM detection, listeners detect the presence of SAM, while in SAM-rate discrimination listeners discriminate between different SAM rates. Here we report three perceptual-learning investigations into the processing that underlies performance on these tasks. In all three studies, a group of naive listeners established a baseline pre-test measure of performance on different SAM-detection or SAM-rate-discrimination conditions. Half of these listeners were trained on one of the pre-test conditions while the rest were untrained controls. After the training phase, all listeners repeated the pre-test conditions to see if there was learning on the trained condition, and if so, whether that learning affected performance on, or generalized to, any of the untrained conditions. We assumed that learning would only generalize if the same processing governed performance on both the trained and untrained condition; in this way we inferred the organization of the processing affected by training. In the first investigation, we examined SAM-rate-discrimination learning and found evidence that, for the training-affected processing, the pitch elicited by SAM noise is processed separately from that of other pitch types, that different modulation rates are largely processed separately, and that the processing underlying rate discrimination is separate from that involved in SAM detection. In the second investigation, we again examined rate-discrimination learning and inferred that training affected a mechanism that processed different carrier spectra of the same BW together; it was uncertain whether stimuli of different bandwidths had shared processing. Finally, we examined SAM-detection learning and concluded that, for the mechanism affected by training, there was shared processing for different modulation rates, separate processing for different carrier spectra, and additional evidence for separate processing of SAM-detection and SAM-rate-discrimination tasks. Taken together, these results suggest that rate-discrimination and detection tasks are governed by separate processing mechanisms, and that the generalization pattern seen in a perceptual-learning experiment is influenced by both the requirements of the trained task and the processing pathway accessed by that task.