Keyword Spotting Call Analysis Without Transcription
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing and classifying voice calls, such as call recording and transcription, face challenges including privacy concerns, high costs, and inaccuracies, especially when dealing with large vocabularies or sensitive information, making it difficult for businesses to effectively assess customer interactions and advertising campaign outcomes.
Innovation Solution
A system that analyzes audio from telephone calls without transcription using a hierarchical keyword vocabulary structure, identifying keywords indicative of specific outcomes through probabilistic models and machine learning techniques, while masking sensitive information to protect privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If call recording and transcription are used to analyze customer interactions, then businesses can gain insight from conversations with real customers, but this incurs high costs, raises privacy concerns, and requires expensive specialized equipment
Solution Approach 1:
The patent extracts only the essential keyword information from calls using automated keyword spotting technology, rather than recording and transcribing entire calls. This selective extraction of critical data points (keywords and phrases) eliminates the need for expensive transcription services while maintaining the ability to analyze customer interactions and determine call outcomes.
Solution Approach 2:
The patent creates simplified representations of call content through keyword metadata tags instead of full transcriptions. These keyword copies capture the essential information needed for analysis without requiring the complete textual reproduction of call content, thereby reducing storage and processing requirements.
2Loss of information
If call recording and transcription are used to track customer interactions, then businesses can understand call outcomes and advertising campaign effectiveness, but this raises legal and privacy concerns regarding sensitive information such as medical health, Social Security Numbers, credit card numbers, and other Personally Identifiable Information
Solution Approach 1:
The patent extracts only keyword-level information from calls without capturing or storing sensitive personal data. By focusing on keyword spotting rather than full transcription, the system obtains call outcome information while inherently excluding personally identifiable information, thus eliminating privacy violation risks associated with storing sensitive customer data.
Solution Approach 2:
The patent uses temporary, disposable keyword metadata tags that are generated during call analysis and then discarded after serving their analytical purpose. These transient keyword representations provide the necessary call outcome understanding without creating persistent records of sensitive information, thereby eliminating long-term privacy risks.
3Loss of information
If transcription of calls is used to analyze customer interactions, then businesses can review call content, but transcription accuracy decreases when the source audio is of poor quality or when large vocabulary speech recognition is required
Solution Approach 1:
The patent extracts keyword information directly from audio signals using specialized keyword spotting algorithms that are optimized for detecting specific predefined terms. This direct extraction approach bypasses the intermediate transcription step entirely, eliminating transcription accuracy problems that arise from poor audio quality or large vocabulary requirements.
Solution Approach 2:
The patent focuses on detecting only the specific keywords and phrases that are relevant to call outcome determination, rather than attempting to transcribe the entire call content. This selective detection approach concentrates computational resources on identifying critical information while ignoring irrelevant content, thereby achieving high accuracy for the specific analytical purpose without requiring full transcription accuracy.
4Loss of information
If recording and transcription of every call is performed to maintain a complete record, then businesses can analyze all customer interactions, but the storage and processing requirements become increasingly prohibitive as call volume increases
Solution Approach 1:
The patent extracts only essential keyword metadata from calls, creating minimal data representations that capture call outcome information. This extraction approach reduces storage requirements from gigabytes per call (full transcription) to kilobytes per call (keyword tags), making it feasible to maintain comprehensive records of all customer interactions even at high call volumes.
Solution Approach 2:
The patent inverts the traditional approach by not storing the full call content and instead storing only the extracted keyword metadata. This inversion of the storage strategy - storing what is essential rather than what is complete - enables scalable storage that can handle increasing call volumes without prohibitive costs.
Data Source
AI summary
A facility and method for analyzing and classifying calls without transcription via keyword spotting is disclosed. The facility uses a group of calls having known outcomes to generate one or more domain- or entity-specific grammars containing keywords and related information that are indicative of particular outcome. The facility monitors telephone calls by determining the domain or entity associated with the call, loading the appropriate grammar or grammars associated with the determined domain or entity, and tracking keywords contained in the loaded grammar or grammars that are spoken during the monitored call, along with additional information. The facility performs a statistical analysis on the tracked keywords and additional information to determine a classification for the monitored telephone call.