Pseudo-Speech Masking for Privacy-Safe Speech Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for securing sensitive content in speech signals, such as Personal Identifiable Information (PII) and Personal Health Information (PHI), in cloud-based automatic speech recognition (ASR) systems result in signal discontinuities due to surrogation without corresponding audio, leading to issues with downstream speech processing systems.
Innovation Solution
The implementation of pseudo-speech representations, which are unintelligible utterances with reduced acoustic-phonetic substance, are generated for sensitive content portions in speech signals, allowing secure processing without disrupting downstream systems by using voice converters and speech processing systems that can decode or ignore these representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If text-level surrogation is used to remove PII from training data, then privacy protection is improved, but signal discontinuities appear in speech signals processed by downstream systems
Solution Approach 1:
The patent introduces pseudo-speech representations as intermediary elements that bridge the gap between privacy protection and signal continuity. These pseudo-speech segments act as mediators that replace PII-containing speech segments while maintaining acoustic signal continuity, allowing downstream systems to process uninterrupted audio streams without exposing sensitive information.
Solution Approach 2:
The patent transforms the nature of speech signals by changing their acoustic parameters in the replaced segments. Instead of using text-level surrogation or artificial speech, the system modifies the audio signal's spectral and temporal characteristics to create pseudo-speech representations that maintain signal continuity while removing PII, effectively changing the signal's physical properties to resolve the contradiction.
2Object-affected harmful factors
If conventional surrogation methods are used, then PII removal is achieved, but downstream speech processing accuracy deteriorates due to signal discontinuities
Solution Approach 1:
Pseudo-speech representations serve as intermediary elements that preserve the acoustic signal's continuity for downstream processing while eliminating PII. These intermediaries maintain the signal's temporal and spectral properties, ensuring that speech processing systems receive uninterrupted audio streams that preserve processing accuracy without exposing sensitive information.
Solution Approach 2:
The system changes the acoustic parameters of replaced segments by generating pseudo-speech representations with modified spectral and temporal characteristics. This parameter transformation maintains signal continuity and preserves the acoustic signal's properties, ensuring downstream speech processing systems receive uninterrupted audio streams that maintain processing accuracy.
3Object-affected harmful factors
If PII is removed from speech signals, then privacy compliance is improved, but the integrity of the speech signal deteriorates
Solution Approach 1:
Pseudo-speech representations function as intermediary elements that maintain the speech signal's structural integrity while removing PII. These intermediaries preserve the signal's temporal continuity, spectral characteristics, and acoustic properties, ensuring that the modified signal remains structurally sound and suitable for downstream processing without compromising privacy.
Solution Approach 2:
The system applies parameter changes to the replaced segments by generating pseudo-speech representations with appropriate spectral and temporal characteristics. This transformation maintains the signal's acoustic properties and structural integrity, ensuring that the modified speech signal remains consistent and suitable for downstream processing while eliminating PII.
Data Source
AI summary
A method, computer program product, and computing system for processing a speech signal. A sensitive portion of the speech signal is identified. A pseudo-speech representation of the sensitive portion is generated using a voice converter system. Speech processing is performed on the speech signal and the pseudo-speech representation of the sensitive portion using a speech processing system.