System and Method for Generating an Audiogram and Application Thereof to Personalise Frequency Response in a User Device

An audio output system including control circuitry configured to be dynamically adaptable to support adjustment of a frequency response of the audio output system through adaptation of processing parameters overseen by the control circuitry, wherein the control circuitry is responsive to data relating to hearing thresholds determined through an on-line networked interaction of the audio output system with a reinforcement learning (RL)-based audiometry system based on a neural network trained using a deep learning reinforcement framework configured with an agent and wherein the neural network is arranged to implement a policy based on temporal difference learning, the RL-based audiometry system (88) arranged to generate an audiogram in which a hearing threshold at a plurality of selectable test frequencies is identified, wherein the system is configured to:

• control generation of selected word-level pairings within which are compound frequencies, the generation of selected word-level pairing dictated by the policy, and wherein each word-level pair represents a current state of the environment;
• effecting recording, in memory, of perception responses to those selectively generated word-level pairings presented from an audio output device;
• and
• identify a hearing threshold for frequencies within the word-level pairings to assemble the audiogram from perception responses for a plurality of different word-level pairings;
• wherein selection, by the agent based on its policy, of each word-level pairing takes into account the current state of the environment, st, together with previous perception responses associated with different word-level pairings and defined parameter data pertaining to personal characteristics.

A hearing evaluation system based on a deep learning reinforcement framework configured with an agent and a neural network arranged to implement a policy and in which evaluation system intelligence is arranged to control overall operation of the evaluation system, wherein the neural network trained on a dataset of historical audiograms augmented by at least demographic data associated with each historical audiogram, and wherein a system intelligence is configured to:

• record a negative reward for each presented state pairing of a plurality of word-level pairings with respective frequency-sound intensity components level;
• and
• following presentation of a first state pairing with a first set of compound test frequencies with associated selected first sound intensity levels, to move to a second different state pairing regardless of an interactive response determined by the system intelligence;
• record, as received by the system intelligence, a first subject response with an association to the first state pairing;
• and
• select and present a second state pairing at a different second test set of compound frequencies and with different associated second sound intensity levels, wherein selection of the second state pairing is based on a probability outcome for the neural network taking into account at least the first subject response;
• oscillate, at least once, between the first and second word-level pairings whilst respectively changing respective sound levels presented therein, thereby to identify a threshold level of hearing through audible resolution across frequency bands;
• and
• generate a visual representation of an audiogram based on reported positive responses to audio detection determined to be at a subject's threshold level of hearing;
• and whereby consequently
• the agent policy is adapted to influence hearing loss evaluation strategy through selection of further state pairings based on responses to current and historical state environments and data related at least one of:
• (a) physiological characteristics of a test subject;
• (b) demographic data for a test subject;
• and
• (c) descriptive information describing a test subject.

A method of evaluating a hearing level threshold, the method comprising:

• training a neural network using a deep learning reinforcement framework set up with an agent and such that the neural network implements a policy based on temporal difference learning;
• generating word-level pairings within which are compound frequencies, the generation of selected word-level pairing dictated by the neural network, each frequency-level pair representing a current state of an environment;
• recording, in memory, perception responses to those selectively generated frequency-level pairings presented from audio output device;
• and
• identifying a hearing threshold frequencies within the word-level pairings to assemble the audiogram from a plurality of perception response at a plurality of different word-level pairings;
• wherein selection of each word-level pairing takes into account the current state of the environment, st, together with previous perception responses associated with different word-level pairings and defined parameter data pertaining to personal characteristics and selection is made by the agent based on its policy;
• and the method further includes:
• generating an audiogram in which a hearing threshold at a plurality of frequencies is identified;
• and
• outputting the audiogram either on a user interface or transmitting data from the audiogram over a network to a device located remotely across the network.