System and method for generating an audiogram and application thereof to personalise frequency response in a user device

Abstract

Audiometry system and method based on a neural network (NN) trained using deep learning reinforcement process 20, with an agent 22 and policy based on temporal difference learning. Frequency-level pairings (FLPs) are selected as dictated by the NN. FLPs are output by an audio device and perception responses recorded. Hearing thresholds for FLPs are identified and an audiogram is assembled. Selection of each FLP accounts for the current state of the environment, previous perception responses for different FLPs and data pertaining to personal characteristics. Optionally, characteristics include physiological characteristics, demographic data or descriptive information for a test subject. The audiogram may be output by a user interface or a remote device. Other embodiments define modifying a frequency response of a speaker by adapting processing parameters responsible for controlling soundwave generation in response to identifying determined hearing thresholds. Further embodiments define estimating an audiogram by training a NN using historical audiograms 27 augmented by demographic data. A first state pairing is presented at a first frequency and sound level, a second selected different state pairing is then presented regardless of a response determined by the system. [Fig. 2]