Phonetic Text Filtering via Speech Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In text-to-speech environments, profanity filters fail to detect verbally offensive content as they are designed to scan text, allowing malicious users to create offensive language through phonetic pronunciations of non-offensive words, which become inappropriate when spoken.
Innovation Solution
An information handling system converts text to synthesized speech and then back to text using speech-to-text conversion, analyzing the resulting text against a filtering criterion to prevent the broadcast of potentially offensive content based on its phonetic pronunciation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard profanity filters are used to scan text, then text processing efficiency is maintained, but phonetically offensive content is not detected
Solution Approach 1:
The patent introduces speech-to-text conversion as an intermediary step between the original text and the profanity filter. The system converts text to speech and then back to text, creating a phonetic representation that the filter can analyze. This intermediary process enables detection of phonetically offensive content without requiring the filter to directly analyze speech signals.
Solution Approach 2:
The filtering system is segmented into multiple independent components: text-to-speech converter, speech-to-text converter, and profanity filter. Each component performs a specific function, allowing the system to handle phonetic analysis separately from the filtering logic. This segmentation enables the system to detect phonetically offensive content while maintaining modularity and ease of maintenance.
2Measurement precision
If text is converted to speech and back to text for analysis, then phonetic offensive content can be detected, but processing time increases
Solution Approach 1:
The system performs preliminary text-to-speech conversion before filtering, creating a phonetic representation in advance. This preliminary action allows the subsequent speech-to-text conversion to focus only on analyzing the phonetic characteristics relevant to profanity detection, rather than processing the entire speech signal from scratch.
3Productivity
If in-place profanity filters are used, then most offensive text is blocked, but creatively crafted phonetic bypasses succeed
Solution Approach 1:
The system creates a phonetic copy of the original text through text-to-speech and speech-to-text conversion. This copy preserves the phonetic characteristics of the original text while allowing the filter to analyze it in a different representation. Malicious content that evades detection in the original text form is revealed in its phonetic copy.
Data Source
AI summary
An approach is provided in which an information handling system converts a first set of text to synthesized speech using a text-to-speech converter. The information handling system then converts the synthesized speech to a second set of text using a speech-to-text converter. In response to converting the synthesized speech to the second set of text, the information handling system analyzes the second set of text against a filtering criterion and prevents usage of the synthesized speech based on the analysis.