Audio Stream Analysis for Human Automated Call Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio processing technologies face challenges in accurately determining whether a phone call is answered by a human or an automated device, as existing methods struggle to differentiate between human and automated responses based on audio patterns, particularly in real-time scenarios involving varying audio streams.
Innovation Solution
A method that performs real-time signal analysis on audio streams to compute a stream audio print, applies a silence constraint to match periods of silence, and computes a total scoring metric function to identify matches between the stream audio print and known audio prints, using tunable parameters to scale components and determine the likelihood of a human or automated response.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If signal analysis is performed on audio streams to compute stream audio print and match with known audio prints, then the accuracy of determining whether a call is answered by human or automated device is improved, but the processing time and computational complexity increase
Solution Approach 1:
The audio stream is divided into multiple segments or frames, and signal analysis is performed on each segment independently to compute audio prints. This segmentation allows parallel processing and reduces the computational burden on the entire stream, thereby decreasing processing time while maintaining accuracy through comprehensive coverage of all audio portions.
Solution Approach 2:
Known audio prints from automated devices are pre-computed and stored in a database before actual call analysis. During real-time call determination, the system only needs to compare incoming audio prints against these pre-computed references, significantly reducing processing time during critical real-time operations while maintaining high accuracy through extensive pre-analysis.
2Reliability
If silence constraint is applied to match periods of silence in audio prints, then the reliability of matching automated device responses is improved, but the complexity of the matching algorithm increases
Solution Approach 1:
Silence periods are extracted and identified as distinct features from the audio stream, separate from spoken content. By isolating silence as a specific characteristic to match against known automated device patterns, the system improves reliability of identification while managing complexity through focused analysis of only the silence portions rather than processing the entire audio stream with full algorithmic complexity.
Solution Approach 2:
The silence constraint is applied locally to specific time periods within the audio stream where silence occurs, rather than uniformly across the entire audio. This allows the matching algorithm to focus computational resources on critical silence regions that are indicative of automated device responses, improving reliability where it matters most while reducing overall algorithmic complexity by not applying full complexity throughout the entire audio processing pipeline.
3Measurement precision
If total scoring metric function is computed by overlaying stream audio print with known audio print at multiple temporal points, then the precision of match identification is improved, but the computational load increases
Solution Approach 1:
Instead of computing the scoring metric function at every single temporal point in the audio stream, the system selects a subset of representative temporal points for computation. This partial action approach maintains precision by analyzing key moments that are most indicative of matches while significantly reducing computational load by avoiding redundant calculations at less critical time points.
Solution Approach 2:
Potential match regions are pre-identified based on preliminary comparison criteria before the full scoring metric function is applied. This allows the system to focus computational resources on only those temporal regions where matches are likely to occur, improving precision where it matters while reducing overall computational load by avoiding exhaustive analysis of the entire audio stream.
Data Source
AI summary
A system receives an audio stream from an application, and performs signal analysis on the audio stream to compute (on sound and silence) a sequence of events. The sequence of events identifies a pattern comprised of at least two of sound, silence, frequency, magnitude, duration, tone, and Dual Tone Multi Frequency (DTMF). The pattern matches a similar pattern of at least two of sound, silence, frequency, magnitude, duration, tone, and DTMF, and the signal analysis is performed on sound and silence within the audio stream. The similar pattern is associated with at least one of an automated device or human answering the call. The system determines whether the sequence of events indicates the phone call is answered by the automated device or the human, based on rules in a heuristic engine and/or a known audio print from a library of known audio prints.