Acoustic Shock Detection Using Two-State Markov Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting acoustic shocks in audio streams, particularly in speech or music, are not robust and often result in false alarms or missed detections, especially when shocks are at low levels or mixed with other signals, leading to degradation of voice quality.
Innovation Solution
A method using a two-state Markov model with a transition matrix and emission matrix to analyze audio frames, assigning category values and determining the probability of acoustic shocks, allowing for better separation between shock and non-shock periods, and independent detection from sound levels, coupled with physical indicators like energy and zero crossing rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If temporal approaches (envelope) and detection of signals of too high energy are used, then detection of high energy shocks is achieved, but it is impossible to distinguish speech signal from shock and detect low energy shocks mixed into other signals
Solution Approach 1:
The audio stream is decomposed into audio frames, and each frame is analyzed independently to assign category values. This segmentation allows the system to process and detect shocks in small manageable units, improving the ability to distinguish shocks from speech signals while reducing false alarms through localized analysis.
Solution Approach 2:
The invention transitions from temporal approaches to a two-state Markov model that operates in the probability domain. By modeling the audio stream as a sequence of states (shock vs. no shock) with transition probabilities, the system achieves better separation between shock and non-shock periods, effectively adding a probabilistic dimension to the detection process.
2Measurement precision
If detailed time-frequency analysis with physical parameters and threshold values is used, then detection of pure tone shocks is improved, but the detection remains not robust to mixed shocks and produces numerous false alarms
Solution Approach 1:
The invention changes the detection parameters from fixed physical thresholds to dynamic transition probabilities in a Markov model. The model uses emission probabilities that adapt to the statistical characteristics of the audio stream, allowing robust detection of mixed shocks without producing excessive false alarms. The category values are discretized into a small number of states, simplifying the parameter space while maintaining detection effectiveness.
Solution Approach 2:
The Markov model inherently uses feedback by considering the sequence of previous frame categories when determining the current state probability. The transition matrix captures the temporal dependencies between frames, allowing the system to learn from past observations and improve current detection decisions, thereby increasing robustness against mixed shocks.
3Reliability
If prior detection systems with multiple floors (limiters, compressors, noise reducers) are used, then protection against acoustic shock is provided, but voice quality is degraded due to filtering of useful speech components
Solution Approach 1:
The invention extracts only the essential detection function from complex multi-floor processing systems. By using a two-state Markov model with category value assignment, the system isolates the shock detection capability from the speech processing chain, enabling protection against acoustic shocks without the need for aggressive filtering that degrades voice quality.
Solution Approach 2:
Instead of applying multiple floors of processing (limiters, compressors, noise reducers), the invention uses a partial action approach with a single Markov model-based detector. This streamlined approach provides sufficient protection against acoustic shocks while avoiding the excessive processing that leads to voice quality degradation.
4Productivity
If WOLA type filter bank is used for pattern search, then shock detection is performed, but the solution remains not very robust and can lead to false shock detections
Solution Approach 1:
The invention replaces the mechanical WOLA filter bank approach with a statistical Markov model. Instead of using fixed filter banks and pattern matching, the system uses probability transitions and emission matrices to detect shocks, achieving both speed and robustness through statistical inference rather than mechanical signal processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method for detecting acoustic shocks in an audio stream, characterized in that it comprises the following steps: breaking down the audio stream into audio frames; analyzing said audio frames in order to assign each audio frame a category value from among a plurality of predefined values; and determining the probability of an acoustic shock occurring in a current frame, based on a sequence of a given length of category values assigned to a set of frames, using a two-state Markov model, defined by a predetermined transition matrix and transmission matrix. The present invention also relates to a device for detecting acoustic shocks.