Audio Call Categorization Using Truncated Transcripts and ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for categorizing audio calls in customer support sessions lack accuracy and efficiency, as they often require processing entire recordings and do not utilize advanced machine learning techniques effectively.
Innovation Solution
A system that truncates audio recordings to a defined length, performs speech recognition, and uses a transformer-based machine learning model to predict categories associated with each call, enabling faster and more accurate categorization by converting audio to text and analyzing transcripts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire audio recording is processed for categorization, then the categorization accuracy is improved, but the processing time increases
Solution Approach 1:
The patent extracts and processes only the most relevant segments of audio calls (such as the beginning and end portions) rather than analyzing entire recordings. This selective extraction maintains categorization accuracy by focusing on segments that contain the most diagnostic information for call outcomes, while significantly reducing processing time and computational resources required.
Solution Approach 2:
The audio call processing is divided into multiple segments (e.g., call opening, call closing, intermediate portions). The system processes only specific segments that are most relevant for categorization purposes, rather than analyzing the complete audio recording continuously. This segmentation approach preserves accuracy for key decision points while reducing overall processing burden.
2Measurement precision
If advanced machine learning models are used for categorization, then the categorization accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent applies machine learning models selectively to processed text transcripts rather than analyzing raw audio directly. By converting audio to text first and then applying NLP-based categorization, the system achieves high accuracy with reduced computational complexity compared to processing entire audio streams through complex audio analysis models.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system achieves more accurate and faster categorization of audio calls by leveraging machine learning models, providing insights and visualizations that enhance customer support analysis.
Implementation Method 1
The third-party transcription service may generate the transcript by converting the audio file to text using a speech-to-text algorithm
Data Source
AI summary
Some embodiments provide a non-transitory machine-readable medium that stores a program. The program receives a set of audio files. Each audio file in the set of audio files includes audio from an audio call. The program further truncates each audio file in the set of audio files to a defined call length. For each audio call in the set of audio calls, the program also receives a transcript of the audio call based on the audio file of the audio call. For each audio call in the set of audio calls, the program further uses the transcript of the audio call as input to a machine learning model for the machine learning model to predict a category from a plurality of categories that is associated with the audio call.