Quantization Table Updating for Voice Situation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Call centers face challenges in detecting specific conversation situations between operators and customers without comprehensive keyword setting, which is time-consuming and requires significant manual effort, and struggles to determine the optimal number of keywords needed for accurate detection.
Innovation Solution
A system using a learning device that performs vector quantization on voice data and generates a quantization table, allowing for the detection of specific conversation situations without pre-defined keyword settings by machine learning an LSTM and DNN model based on correct answer information, enabling accurate determination of abnormal conversation situations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If comprehensive keyword setting is used to detect specific conversation situations, then detection accuracy is improved, but manual effort and time consumption increase significantly
Solution Approach 1:
The system performs self-service by automatically learning and updating the quantization table using accumulated conversation data without requiring manual keyword setting. The learning device autonomously identifies important conversation patterns and updates the quantization table accordingly, eliminating the need for continuous manual intervention while maintaining high detection accuracy.
Solution Approach 2:
The quantization table is pre-generated and stored in advance using accumulated conversation data from multiple operators and customers. This preliminary action allows the system to be ready for immediate use without requiring time-consuming keyword setting when deployment is needed, thus reducing initial setup time while maintaining detection accuracy.
2Measurement precision
If comprehensive keyword setting is used to detect specific conversation situations, then detection accuracy is improved, but device complexity increases
Solution Approach 1:
The patent replaces the mechanical keyword-matching system with a learning-based acoustic model system. Instead of manually configuring keywords and rules, the system uses vector quantization and neural network models (LSTM and DNN) to automatically learn conversation patterns from data, significantly reducing system complexity while maintaining or improving detection accuracy.
Solution Approach 2:
The system changes the fundamental parameters of conversation detection from keyword-based discrete values to continuous acoustic features that are quantized into vectors. This parameter transformation allows the system to capture nuanced conversation situations without requiring extensive keyword configuration, thereby reducing complexity while improving detection capability.
3Measurement precision
If the quantization table is updated frequently to improve detection accuracy, then measurement precision is improved, but processing time increases
Solution Approach 1:
The learning device performs periodic updates of the quantization table at predetermined intervals rather than continuously or frequently. This periodic action allows the system to maintain detection accuracy by updating with accumulated data while avoiding excessive processing time associated with frequent updates. The interval is optimized to balance accuracy improvement with processing efficiency.
Data Source
AI summary
A non-transitory computer-readable recording medium having stored therein an update program that causes a computer to execute a procedure, the procedure includes calculating a selection rate of each of a plurality of quantization points included in a quantization table, based on quantization data obtained by quantizing features of a plurality of utterance data, and updating the quantization table by updating the plurality of quantization points based on the selection rate.