Voice Activity Detection on Coded Speech Bitstream Without Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice activity detection (VAD) methods in digital networks require decoding speech signals, which is computationally expensive and not robust to noise, especially when using conservative codecs like AMR.
Innovation Solution
A system that performs VAD directly on the coded bitstream using a parameter extraction module and a classifier like CART or DBN, without decoding, to classify frames based on bitstream coding parameters, with optional smoothing and hysteresis modules to refine decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech signals are decoded to perform VAD, then VAD accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent extracts only the necessary coding parameters from the compressed bitstream without performing full decoding. The parameter extraction module selectively retrieves features like spectral coefficients and coding parameters that are sufficient for VAD decisions, avoiding the computationally expensive inverse transform and reconstruction steps that would be required for full decoding.
Solution Approach 2:
The VAD system is segmented into distinct functional modules: a parameter extraction module that operates on the compressed bitstream, a classification module that performs VAD decisions, and optional smoothing modules. This segmentation allows the system to perform VAD on compressed data without requiring full signal reconstruction, reducing computational complexity while maintaining accuracy.
2Reliability
If conservative VAD settings are used in AMR codec, then false positives are reduced, but noise robustness deteriorates
Solution Approach 1:
The patent implements smoothing modules (temporal smoothing and hysteresis) that provide feedback mechanisms to refine VAD decisions. These modules analyze patterns across multiple frames and adjust decisions based on temporal consistency, allowing the system to be more aggressive in detecting speech while maintaining low false positive rates through contextual validation.
Solution Approach 2:
The system dynamically adjusts VAD sensitivity based on operating conditions and can select from multiple classifier configurations. The ability to adapt the classification threshold and smoothing parameters allows the system to optimize between noise robustness and false positive control for different acoustic environments and speech scenarios.
3Measurement precision
If full decoding is performed for VAD, then feature extraction accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary extraction of relevant features directly from the compressed bitstream before any VAD classification occurs. By pre-extracting spectral coefficients and coding parameters in their compressed form, the system avoids the time-consuming full decoding process while still obtaining sufficient feature accuracy for reliable VAD decisions.
Data Source
AI summary
A system, method and computer program product are described for voice activity detection (VAD) within a digitally encoded bitstream. A parameter extraction module is configured to extract parameters from a sequence of coded frames from a digitally encoded bitstream containing speech. A VAD classifier is configured to operate with input of the digitally encoded bitstream to evaluate each coded frame based on bitstream coding parameter classification features to output a VAD decision indicative of whether or not speech is present in one or more of the coded frames.