Central Speech Model Server Using Anonymous Audio for Voice Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Legal regulations prohibit the indefinite storage of audio files used for training voice recognition models, leading to the potential loss of learned speech patterns, which compromises the effectiveness of voice-controlled devices.

Innovation Solution

A central speech model server generates anonymous secondary audio files from randomly selected groups of primary audio files, training the model with these instead of the original files to comply with legal regulations while preserving learned speech patterns.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If audio files are stored indefinitely for training speech models, then the speech model's recognition accuracy improves, but legal regulations are violated

Engineering Contradiction:
Improvespeech recognition accuracyVSAvoidlegal compliance
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent creates synthetic audio files that copy the essential speech pattern characteristics from original user audio files without storing the originals. The synthesis module generates artificial training data that preserves speech pattern information while eliminating legal compliance issues associated with storing actual user recordings.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms the storage parameter from keeping original audio files to storing synthesized audio files. This parameter change maintains the training value for speech pattern recognition while resolving the legal contradiction by changing the nature of the stored data from actual user recordings to synthetic representations.

Inventive Principle:
Principle #35Parameter changes

2Object-affected harmful factors

If audio files are deleted after maximum storage period, then legal regulations are complied with, but the speech model forgets learned speech patterns

Engineering Contradiction:
Improvelegal complianceVSAvoidspeech pattern information
Core Design Contradiction:
Object-affected harmful factorsVSLoss of information

Solution Approach 1:

Instead of deleting original audio files and losing speech pattern information, the system creates synthetic copies that capture the essential speech patterns. These synthetic files preserve the information needed for model training while allowing compliance with data retention regulations.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The synthesis module acts as an intermediary between the original audio files and the speech model training process. It transforms original recordings into synthetic training data, serving as a mediator that preserves speech pattern information while enabling legal compliance through data anonymization.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If more user audio files are collected for training, then the universality of the speech model improves, but user privacy risks increase

Engineering Contradiction:
Improvespeech model universalityVSAvoiduser privacy risk
Core Design Contradiction:
Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

The system synthesizes training audio files from a limited set of original user recordings, creating diverse training data without needing to collect and store numerous actual user audio files. This approach maintains model universality while minimizing privacy risks by working with synthesized rather than actual user data.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The audio synthesis module serves as a privacy-protecting intermediary that enables the system to benefit from user speech patterns without directly storing or processing large quantities of personal audio data. It mediates between the need for diverse training data and user privacy protection.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4546334B1Central language model server for voice-enabled terminal and operating method for central language model server
Publication Date: 2026.03.11 DEUTSCHE TELEKOM AG
  • EP4546334B1 patent drawingFigure 1

AI summary

Method for operating a central speech model server for a voice-controlled terminal, wherein a voice-controlled terminal captures a voice command from a user of the voice-controlled terminal and transmits a primary audio file containing the captured voice command to a central speech model server associated with the voice-controlled terminal, a voice control of the voice-controlled terminal recognizes the voice command in the provided primary audio file by means of a speech model received from the central speech model server and causes the terminal to react accordingly to the recognized voice command, as well as central speech model server for a voice-controlled terminal and computer program product.