Natural-Language Hearing Device Fitting With Generative AI

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing hearing devices struggle to optimally adjust settings in response to changing environments and user preferences without professional intervention, making it difficult to ensure effective hearing enhancement.

Innovation Solution

A generative large language model processes natural language inputs from users to determine embedding vectors, matching them with predefined solutions for hearing device adjustments, allowing users to make real-time adjustments and updates based on user feedback.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If hearing devices are equipped with increasingly sophisticated settings to account for individual hearing loss and environmental conditions, then the hearing enhancement capability is improved, but it becomes less predictable whether the desired goal of optimizing hearing is met

Engineering Contradiction:
Improvehearing enhancement effectivenessVSAvoidnumber of settings
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system enables users to independently adjust hearing device settings by processing their natural language feedback through a generative LLM. The LLM analyzes user inputs, determines appropriate setting adjustments, and implements changes without requiring healthcare professional intervention, allowing the device to serve itself in optimizing hearing settings

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system continuously collects user feedback through natural language inputs and uses this feedback to iteratively adjust settings. The generative LLM processes user feedback, compares current settings with desired outcomes, and makes real-time adjustments to improve hearing enhancement effectiveness based on actual user experience

Inventive Principle:
Principle #23Feedback

2Ease of operation

If users can adjust settings independently without healthcare professional visits, then ease of operation is improved, but measurement precision of hearing issues may deteriorate

Engineering Contradiction:
Improveuser independence in adjustmentVSAvoidaccuracy of hearing issue diagnosis
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The system replaces the mechanical system of professional fitting with an AI-based generative LLM that processes natural language inputs. The LLM acts as an intelligent intermediary, translating user descriptions of hearing issues into precise setting adjustments, thereby maintaining measurement precision while enabling user independence

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The generative LLM serves as an intermediary between the user's natural language feedback and the hearing device settings. It processes user inputs, interprets hearing issues, determines appropriate adjustments, and implements changes, bridging the gap between user experience and technical configuration without requiring professional intervention

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4633203A1Method and apparatus for using a generative large language model for audiological fitting and device control
Publication Date: 2025.10.15 SONOVA AG
  • EP4633203A1 patent drawingFigure 1~2
  • EP4633203A1 patent drawingFigure 3~4
  • EP4633203A1 patent drawingFigure 5

AI summary

A fitting apparatus for a hearing device includes a memory, the memory having stored thereon a plurality of embedding vectors and related solutions for possible hearing issues with a hearing or computing device. The fitting apparatus is configured to: - receive, via a user interface, a first signal indicative of a natural language input indicative of an issue experienced by a user; - determine an embedding based on the natural language input; - generate, based on the embedding, a first embedding vector; - determine a closest match between the first embedding vector and an embedding vector of the plurality of embedding vectors; and - send, to the user interface, a second signal indicative of a reply output, the reply output comprising information related to a solution of the related solutions for the issue experienced by the user.