Phoneme Sequence Analysis for Multi-Language Phrase Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing contact center audio recordings are time-consuming and inefficient, particularly when searching for specific phrases, as they fail to account for regional pronunciation, cadence differences, and homologues, leading to false positives and difficulties in multi-language environments.
Innovation Solution
A multi-language system and method that uses an audio analysis engine to identify and mark specific phrases within spoken audio recordings by comparing phonemes, adjusting for regional accents and cadence, and allowing multiple audio representations of a phrase to be associated with a single identifier, enabling accurate capture and visualization of phrase usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If keyword recognition is used to search for phrases in audio recordings, then the analysis process is accelerated, but false positives increase due to inability to account for regional pronunciation and cadence differences
Solution Approach 1:
The patent segments phrases into individual phonemes and creates phoneme sequences for comparison. This segmentation allows the system to analyze speech at the phoneme level rather than relying on whole-word keyword matching, enabling accurate identification of phrases despite variations in pronunciation, cadence, and regional accents. The phoneme sequence comparison method breaks down the phrase recognition problem into smaller, more manageable units that can be systematically compared against the audio recording.
2Measurement precision
If manual review of entire audio recordings is performed to ensure accurate phrase identification, then measurement precision is maintained, but loss of time increases significantly
Solution Approach 1:
The patent replaces the mechanical process of manual audio review with an automated computer-based phoneme sequence comparison system. The system automatically converts audio recordings into phoneme sequences and compares them against target phrase phoneme sequences, eliminating the need for manual listening and review. This substitution maintains high accuracy through systematic phoneme-level analysis while dramatically reducing the time required for phrase identification.
3Device complexity
If single-language keyword search is used, then device complexity is minimized, but adaptability decreases when handling multi-language and multi-accent environments
Solution Approach 1:
The patent creates a universal phoneme-based search system that can handle multiple languages and accents through a common phoneme sequence comparison framework. Rather than creating separate search systems for each language, the system uses phoneme sequences as a universal representation that works across different languages and regional pronunciations. This universal approach allows the same core algorithm to identify phrases in various languages and accents without requiring language-specific customization.
Data Source
AI summary
A system and method for multi-language phrase identification within spoken interaction audio capable of adjusting for regional pronunciation (accents), cadence differences, and homologs. In this system, a spoken interaction audio data store supplies spoken audio data such as contact center call recordings to be analyzed for a specific phrase or set of phrases. Phrases are entered as natural language text and converted to the phonemes representative of the phrase audio using the invention's language packs and stored in a data store. Spoken interaction and phrase audio are converted to a digital format allowing comparison using multiple characteristics. Phrase matches are stored for subsequent post analysis display and analytics generation.