Pronunciation Map Generation for Speech-to-Text Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech-to-text and text-to-speech systems often mispronounce words, especially names, due to their reliance on prebuilt dictionaries that ignore phonetic rules of the native language and fail to incorporate phonetics of the original speech language, limiting pronunciation suggestions.
Innovation Solution
A method and system that analyze a sound file to determine its language and the display language, create a pronunciation map using phonological rules, and generate pronunciation hints that incorporate the phonetics of the original speech, allowing for accurate pronunciation in the user's dialect, with optional graphical or image aids.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If prebuilt dictionaries are used for speech-to-text and text-to-speech, then system simplicity is maintained, but pronunciation accuracy deteriorates
Solution Approach 1:
The system segments the pronunciation process into multiple components: phoneme extraction from sound files, phonological rule application, and pronunciation hint generation. This segmentation allows the system to maintain simplicity while improving accuracy by handling each aspect separately and systematically.
Solution Approach 2:
The system performs preliminary analysis of sound files to extract phonemes and create pronunciation maps before generating text-to-speech output. This preliminary action enables the system to incorporate phonetic rules from the original speech language, improving pronunciation accuracy without complicating the overall system architecture.
2Productivity
If prebuilt dictionaries are used, then processing speed is maintained, but pronunciation reliability deteriorates
Solution Approach 1:
The system creates pronunciation maps and extracts phonological rules from sound files in advance, before the actual speech-to-text or text-to-speech conversion is needed. This preliminary action stores phonetic information that can be quickly retrieved during processing, maintaining speed while improving reliability.
Solution Approach 2:
The system automatically extracts phonemes from sound files and generates pronunciation hints without requiring manual dictionary curation. This self-service approach improves reliability by capturing actual pronunciation patterns from native speech while avoiding the time-consuming process of manual dictionary creation and maintenance.
3Manufacturing precision
If phonological rules and phonetics from original speech are incorporated, then pronunciation accuracy is improved, but system complexity increases
Solution Approach 1:
The system introduces phoneme extraction and phonological rule analysis as intermediary steps between the sound file and the text-to-speech conversion. These intermediaries process the phonetic information and transform it into a format that can be used by the existing speech synthesis system, improving accuracy without requiring complete system redesign.
Solution Approach 2:
The system creates pronunciation maps that copy and store the phonetic characteristics of native speech. These maps serve as reference data that can be applied during text-to-speech conversion, allowing the system to replicate authentic pronunciation patterns without needing to reprocess the original sound files each time.
4Reliability
If phoneme extraction from sound files is performed, then pronunciation reliability is improved, but processing time increases
Solution Approach 1:
The system extracts phonemes from sound files and creates pronunciation maps in advance, before the actual speech conversion task is performed. This preliminary processing stores the phonetic information in a reusable format, so that subsequent text-to-speech conversions can quickly reference the pre-extracted phonemes without reprocessing the original audio, reducing time loss while maintaining reliability.
Data Source
AI summary
In an approach to analyzing a sound file, determining the language of the sound file and the display, creating a pronunciation map between the languages, generating a set of pronunciation hints based on the pronunciation map, and displaying the set of pronunciation hints, one or more computer processors identify a word from one or more words in a sound file. The one or more computer processors determine a dialect of spoken language for the word. The one or more computer processors determine a different language to display the word. The one or more computer processors retrieve one or more phonological rules based on the determined spoken language of the word and the determined different language to display the word. The one or more computer processors create a pronunciation map based on the retrieved phonological rules of the word.