Speech Codec Prediction Switching for Lower LTP Bit Consumption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In speech coding, the Long-Term Prediction (LTP) processing consumes too many bits when the speech signal is less periodic, leading to reduced compression performance due to unnecessary inclusion of the LTP residual signal in the bit stream.
Innovation Solution
A method and apparatus for adaptive encoding and decoding that determine a long-term flag value to decide whether to perform LTP processing, allowing the encoding process to skip LTP processing when the flag is set to a second value, thus improving compression performance by omitting unnecessary LTP contribution signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If LTP processing is always performed for speech signals, then the tone structure of voiced speech is preserved, but the bit consumption increases when the speech signal is less periodic
Solution Approach 1:
The patent applies dynamics by making the LTP processing adaptive rather than static. The encoder determines a long-term flag value based on the periodicity of the speech signal, and the decoder adaptively switches between performing LTP processing (when flag indicates periodic speech) and skipping it (when flag indicates aperiodic speech). This dynamic adaptation resolves the contradiction by preserving tone structure only when necessary while reducing bit consumption when LTP would be ineffective.
Solution Approach 2:
The patent changes the parameter of LTP processing application from a fixed state to a variable state based on speech periodicity. By introducing the long-term flag parameter that reflects the periodicity characteristic of the speech signal, the system adjusts whether LTP processing is applied. This parameter change enables the system to maintain reliability for voiced speech while minimizing bit consumption for unvoiced or aperiodic segments.
2Measurement precision
If LTP residual signal is always included in the bit stream, then the decoding accuracy is maintained, but the compression performance is reduced
Solution Approach 1:
The patent applies the extraction principle by selectively removing the LTP residual signal from the bit stream when the speech signal is determined to be less periodic. The encoder extracts only the necessary components (LTP residual signal) when they contribute to decoding accuracy, and omits them when they would waste bits without improving quality. This selective extraction resolves the contradiction between maintaining decoding accuracy and achieving compression performance.
3Reliability
If LTP processing is performed for all speech signals, then the long-term dependence is removed, but the encoding complexity increases for aperiodic signals
Solution Approach 1:
The patent applies partial action by performing LTP processing only partially - specifically, only when the speech signal exhibits periodic characteristics. The encoder uses the long-term flag to determine whether to apply LTP processing, thereby avoiding the excessive action of applying it to all speech signals regardless of their periodicity. This partial application reduces encoding complexity for aperiodic signals while maintaining reliability for periodic signals.
Data Source
AI summary
The present disclosure relates to a method, apparatus, and system for encoding and decoding signals. The encoding method includes: converting a first-domain signal into a second-domain signal; performing Linear Prediction (LP) processing and Long-Term Prediction (LTP) processing for the second-domain signal; obtaining a long-term flag value according to a decision criterion; obtaining a second-domain predictive signal according to the LP processing result and the LTP processing result when the long-term flag value is a first value; obtaining a second-domain predictive signal according to the LP processing result when the long-term flag value is a second value; converting the second-domain predictive signal into a first-domain predictive signal, and calculating a first-domain predictive residual signal; and outputting a bit stream that includes the first-domain predictive residual signal.