Cumulative Spectral Entropy for Tone Detection in Degraded Calls
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing call progress analysis systems struggle to accurately differentiate between tones and speech in audio signals, particularly in degraded network conditions, due to the lack of standardized tone frequencies and patterns across telephony providers, leading to insufficient performance.
Innovation Solution
Implementing cumulative average spectral entropy analysis to distinguish tones from speech by calculating cumulative average power spectral amplitude and entropy, using a difference measure to classify audio signals, thereby improving tone detection in degraded audio conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional signal processing algorithms are used for tone detection, then the system is simple to implement, but the classification accuracy deteriorates in degraded audio conditions
Solution Approach 1:
The patent transforms the audio signal from time-domain to frequency-domain representation using Short-Time Fourier Transform (STFT), changing the parameter space to enable spectral entropy calculation. This parameter transformation allows the system to capture frequency distribution characteristics that are invisible in the time domain, thereby improving tone-speech differentiation accuracy without requiring complex hardware modifications
Solution Approach 2:
The patent introduces spectral entropy as an intermediary metric that mediates between the raw audio signal and the final classification decision. By calculating the entropy of the power spectral density, the system creates an intermediate representation that quantifies the randomness or predictability of frequency components, serving as a reliable discriminator between tones (low entropy) and speech (high entropy) even in degraded conditions
2Reliability
If cumulative average spectral entropy analysis is implemented, then the classification accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary action by calculating the cumulative average of spectral entropy values over multiple time frames before making the final classification decision. This preliminary averaging process smooths out transient fluctuations and noise in the spectral entropy measurements, creating a more stable and reliable feature for tone-speech differentiation that maintains robustness in degraded network conditions
Solution Approach 2:
The patent applies partial action by focusing computational resources on calculating only the spectral entropy of the power spectral density, rather than analyzing all possible signal characteristics. This selective approach computes only the most discriminative feature (spectral entropy) needed for tone-speech classification, reducing unnecessary computational overhead while maintaining high classification accuracy
3Measurement precision
If standardized tone frequencies are established worldwide, then the detection precision improves, but the adaptability to different telephony providers deteriorates
Solution Approach 1:
The patent implements universality by designing a classification system based on spectral entropy characteristics rather than provider-specific frequency standards. The method universally applies to any telephony provider by analyzing the fundamental difference in spectral entropy between tones and speech, making the system adaptable to various tone frequencies and patterns without requiring provider-specific configuration or recalibration
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
A contact center system for performing call progress analysis including tone and speech classification according to an embodiment includes at least one processor and at least one memory comprising a plurality of instructions stored thereon that, in response to execution by the at least one processor, causes the contact center system to determine a cumulative average of an entropy of an audio signal received by the contact center system, determine a cumulative average power spectral amplitude and a cumulative average spectral entropy of the audio signal, calculate a difference measure of the audio signal as a difference between the cumulative average of the entropy and the cumulative average spectral entropy, distinguish tones from speech of the audio signal based on the difference measure, and process one or more tones of the audio signal in response to identification of the one or more tones in the audio signal.