Anonymized Speech Recognition via Frequency Domain Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in complying with regulatory requirements, such as COPPA, which consider child speech data as personal information, necessitating rigorous consent methods when sharing with third parties, while also needing to perform interactive operations without capturing and storing raw audio files.
Innovation Solution
A method for anonymizing speech data by converting raw waveforms into frequency components and removing identifying features, allowing for speech recognition without retaining raw audio files, thus enabling compliance with less stringent consent requirements and protecting user privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If raw audio files are captured and stored for speech recognition, then speech recognition functionality is maintained, but user privacy protection deteriorates and regulatory compliance becomes difficult
Solution Approach 1:
The patent extracts and removes identifying features from speech waveforms by converting to frequency domain and eliminating specific frequency components that carry speaker identification information while preserving speech content, thereby maintaining speech recognition functionality while protecting user privacy
Solution Approach 2:
The speech waveform is segmented into frequency components through Fourier transform, allowing selective removal of identifying frequency components while retaining speech recognition-relevant features, thus separating privacy-protection-critical elements from functionality-critical elements
2Measurement precision
If speech data is shared with third parties for processing, then speech recognition accuracy is improved, but regulatory compliance requirements become more stringent
Solution Approach 1:
The system performs preliminary anonymization processing on speech waveforms before sharing with third parties, converting to frequency domain and removing identifying features in advance, so that third parties receive only anonymized data requiring milder consent under COPPA regulations
Solution Approach 2:
The patent introduces an intermediary anonymization processing step between speech capture and third-party sharing, using frequency domain transformation and selective component removal as a mediator to decouple speech recognition accuracy from privacy exposure risks
3Object-affected harmful factors
If identifying features are removed from speech waveforms, then user privacy protection is improved, but speech data utility for analysis deteriorates
Solution Approach 1:
The patent applies local quality modification by selectively removing only specific frequency components that carry speaker identification information while preserving other frequency components that contain speech content, achieving differential treatment of waveform components based on their functional importance
Data Source
AI summary
Various of the disclosed embodiments relate to systems and methods for extracting audio information, e.g. a textual description of speech, from a speech recording while retaining the anonymity of the speaker. In certain embodiments, a third party may perform various aspects of the anonymization and speech processing. Certain embodiments facilitate anonymization in compliance with various legislative requirements even when third parties are involved.